《贝恩公司:2023年技术报告(英文版)(84页).pdf》由会员分享,可在线阅读,更多相关《贝恩公司:2023年技术报告(英文版)(84页).pdf(84页珍藏版)》请在三个皮匠报告上搜索。
1、Technology Report 2023 Reset and Reinvent:The Thriving Landscape of Tech Innovation This work is based on secondary market research,analysis of financial information available or provided to Bain&Company and a range of interviews with industry participants.Bain&Company has not independently verified
2、 any such information provided or available to Bain and makes no representation or warranty,express or implied,that such information is accurate or complete.Projected market and financial information,analyses and conclusions contained herein are based on the information described above and on Bain&C
3、ompanys judgment,and should not be construed as definitive forecasts or guarantees of future performance or results.The information and analysis herein does not constitute advice of any kind,is not intended to be used for investment purposes,and neither Bain&Company nor any of its subsidiaries or th
4、eir respective officers,directors,shareholders,employees or agents accept any responsibility or liability with respect to the use of or reliance on any information or analysis contained in this document.This work is copyright Bain&Company and may not be published,transmitted,broadcast,copied,reprodu
5、ced or reprinted in whole or in part without the explicit written permission of Bain&Company.Net Promoter,NPS,NPS Prism,and the NPS-related emoticons are registered trademarks,and Net Promoter Scoresm,NPSxsm,and Net Promoter Systemsm are service marks of Bain&Company,Inc.,NICE Systems,Inc.,and Fred
6、Reichheld.Copyright 2023 Bain&Company,Inc.All rights reserved.AcknowledgmentsDavid Crawford,leader of Bain&Companys Global Technology practice,and a team led by Dana Aulanier,practice director of the Technology Services practice,prepared this report.Bain Partners John Beaumont,Bryce Crawford,David C
7、rawford,Matthew Crupi,Rishi Dave,David Deming,Frank Ford,Jonathan Frick,Arun Ganti,Peter Guarraia,Peter Hanbury,Karen Harris,Simon Heap,Josh Hinkel,Jens Friis Hjortegaard,Anne Hoecker,Christopher Johnson,Jordan Lee,Brian Kmet,David Lipman,Justin Murphy,Rohan Narayen,Brendan ORourke,Christopher Perry
8、,Bill Radzevych,Gene Rapoport,Michael Schallehn,Christopher Schorling,Jen Smith,Balaji Thirumalai,JF Van Kerckhove,Colleen von Eckartsberg,and Jue Wang;Expert Partners Syed Ali,Sanjin Bicanic,Bala Parameshwaran,Ted Shelton,and Velu Sinha;Associate Partners James Baird,Jay Bhatnagar,Arjun Dutt,Peter
9、Henle,and Nihar Naik;Senior Manager Neha Verma;Vice President,Data Science Eric Sheng;and Chief Executive of Enterprise Blueprints Neil Mulholland wrote its chapters.The authors wish to thank Bain Partners Adam Borchert,Bhavi Mehta,and Paul Renno;Associate Partner Tanvee Rao;Senior Managers James Ca
10、rlson and Ruchi Singh;Advisory Partner Mike McKay;Managers Yash Damania,Aditya Poonia,and Collin Skousen;Consultants James Adams,Gavin Cosgrave,Caro Gonzalez,Chris Ma,Deeksha Manjunath,and Dubem Mbeledogu;Associate Consultants Ramya Ajjarapu,Kush Amin,Joshua Chiang,Helena Epstein,Misty Liao,Myron Ma
11、geswaran,Alex Nielsen,and Kit Wiggin;Practice Directors Dana Aulanier,Lauren Brom,and Alex Smyth;Practice Senior Managers Elisabeth Duffy and Tarun Gupta;Associate Directors of Primary Research Mike Kates and Dawn Kix;Leads of Primary Research JP Liss and Christopher Marguette;Bain Capability Networ
12、k Senior Managers Saurabh Gupta and Nishat Parveen;and John Campbell,Jeff Bauter Engel,Adam Jones,and David Sims for their editorial support.1Technology Report 2023ContentsReset and Reinvent:The Thriving Landscape of Tech Innovation .2Value Evolution.5Creating Value in Tech Throughout the Life Cycle
13、 .6Preparing for Exit:A Buyers Market Is Coming for Tech Assets .10AI Investors:Act Fast,Act Wisely .15Technology Enters Its Post-Globalization Era.20Strategic Battlegrounds.26Youre Out of Time to Wait and See on AI .27How AI Is Recoding the Software Business Model .32Generative AI and Cybersecurity
14、:Strengthening Both Defenses and Threats .37Taking the Hyperbole Out of the Metaverse .41The Untapped Value at the Intelligent Edge .47After the Chip Shortage,Fears of a Capacity Glut Are Overblown .52Operational Transformations.58Digital Innovation:Getting the Architecture Foundations Right .59The
15、Talent Implications of Generative AI .63How Enterprise Sales Can Supercharge Product-Led Growth.68How Your Revenue Can Grow Faster Than Your Salesforce.73Decarbonizing Technology Supply Chains .762 2Reset and Reinvent:The Thriving Landscape of Tech InnovationBy David CrawfordTechnology thrives on in
16、novation,and today,the sector is in a period of renewal and reinvention.Tech is healthy,vibrant,and moving forward assertively,driven by strong global demand and broad interest in new technologies.What looked like a slowdown was actually a reset.After the Covid-19 spending bubble,demand slowed,valua
17、tions fell,and profitability mattered again.The sector responded,cutting costs and laying off thousands,then bounced back quickly with market caps near peak levels again(Figure 1).This rebound rides on optimism.Technology is increasingly central to everything we dopersonally,professionally,and as a
18、society.Even during periods of economic uncertainty,companies continue to invest in technology(Figure 2).Large language models and generative AI represent the most significant disruption since the Internet.The need for resilience in supply chains is creating capacity in new places.These are the inve
19、stments that will propel and reshape the tech sector in the years ahead.David Crawford Leader of Bains Global Technology Practice3Technology Report 2023Source:Bain IT Decision Makers Survey(July 2021 n=150;August 2022 n=200;August 2023 n=151)Percentage of CIOs who say that their budgets are flat or
20、increasedAugust 202376August 202280July 202187%Note:Based on Wednesday closing pricesSources:S&P 1500 composite sector data;Bain&CompanyS&P 1500 aggregate market cap,weekly percentage change50300%Worst-performing sectorNext-bestperforming sector TechJanuary 1,2019January 7,2020January 5,2
21、021 January 4,2022January 3,2023Figure 2:Most CIOs still arent decreasing their budgetsFigure 1:The technology sector bounces backCreating Value in Tech Throughout the Life Cycle .6Preparing for Exit:A Buyers Market Is Coming for Tech Assets .10AI Investors:Act Fast,Act Wisely .15Technology Enters I
22、ts Post-Globalization Era.20Value Evolution6Value EvolutionCreating Value in Tech Throughout the Life CycleDriving returns depends on how far down the road you are.By David Crawford,Matthew Crupi,and JF Van KerckhoveAt a Glance In the tech sector,investors are attracted to young,disruptive companies
23、 based on their growth potential.As companies and their markets mature,investors expect a mix of growth and returns.Mature companies with a proven track record in stable markets can expect slower growth while their investors are closely focused on profitability and return on invested capital.Underst
24、anding where your company is in its life cycle and signaling expectations to the right group of investors is essential for attracting the capital that companies need to operate.Over the past five years,as technology companies have navigated through volatile supply-and-demand shocks(Covid-19,supply c
25、hain constraints,recession,inflation)and uncertainty on financing(higher cost of capital,closed equity markets),many have focused on growth at any cost.Their attention is shaped by the common belief that,in tech,growth matters most in creating value for shareholders.7Technology Report 2023The truth
26、is more nuanced:Shareholders reward tech companies differently based on a companys context and point in the life cycle.For more mature companies,their return on invested capital(ROIC)can matter much more.As markets mature,they also attract more conservative investors,who then pressure companies to t
27、urn toward more predictable,profitable opportunities.This then limits further growth and investment in the business and attracts a new breed of value investor who cares about stable revenue streams,reinforcing this cycle.As markets mature,they also attract more conservative investors,who then pressu
28、re companies to turn toward more predictable,profitable opportunities.When determining a companys value,investors often turn to total shareholder return(TSR),a financial metric that indicates the total amount an investor reaps for an investment.TSR has a few fundamental factors that measure strategy
29、 and operations(for example,revenue growth,profitability,and capital efficiency);investor sentiment(multiples);and capital structure(debt-to-equity ratio,buybacks,and dividends).The weight of these factors differs as companies and their markets pass through life stages with different growth rates(se
30、e Figure 1).Early-growth markets may be small,but they are growing fast,disrupting existing industries or creating new ones.Speed and innovation are essential for related businesses as they establish their market position.Revenue growth and future prospects drive TSR,creating value for venture inves
31、tors who invest to scale new champions.Late-growth markets still benefit from robust growth as they expand and stabilize.TSR is based on growth,with credit for initial profitability as investors look for growing companies with proven economic viability and scalable profitability.Mature markets are r
32、elatively stable,with single-digit growth expectations.An incumbents share position may vary by only a few basis points a year.Investors focus on profitability and ROIC.Growth is less important,valued only if its efficiently funded.Asset-rich legacy markets may be fighting the gravitational pull of
33、shifting customer needs.Investors expect highly predictable returns with low to no tolerance for risk.Once a company is within a given stage of market maturity,it is important to maximize the sources of TSR specific to that phase.Although these phases seem logical,too many companies in late-growth a
34、nd mature markets downplay the importance of profitability and ROIC.Growth still matters,but profitability and capital efficiency may create more value.8Technology Report 2023Note:DEC start date is year of first reported annual loss,and end date is year of Compaq acquisitionSources:S&P Capital IQ;Ba
35、in analysisDetractors from TSRContributors to TSRMarket maturity Historic total shareholder return(TSR)catalystsEarly growthLate growthMatureAsset-rich legacyProfitability growthMultiple growthRevenue growthDivided payoutShare countTSR based on futuregrowth prospectsStrategySell shares and build deb
36、t to fund rapid growthExamplesPalo Alto Networks20102022Palantir 20182022CrowdStrike 20172022Snowflake 20192022TSR based on growth and with credit for initial profitabilityStrategyBalance profitability and reinvestmentExamplesAmazon Web Services20112015Microsoft20082015NetApp20102019TSR based on pro
37、fitability growth and capital returnsStrategyDrive operational excellence and ROIC in core markets,returning excess cash to shareholdersExamplesCisco20122022Broadcom20122022Oracle20092022TSR based on capital returns to shareholdersStrategyManage assets to deliver predictable returnsExamplesDigital E
38、quipment Corporation(DEC)19911998Figure 1:The catalysts for total shareholder returns differ according to a companys stage in the life cycle9Technology Report 2023Companies in late-growth and mature stages should be improving the efficiency of generating revenue by:rethinking which markets are most
39、attractive;reducing portfolio complexity and adopting design-to-value principles to match customer needs;revisiting ways of working to improve productivity(including AI and automation);deploying new business models based on asset-light solutions;pursuing M&A where operational synergies are clear;and
40、 reinvigorating a lean core to ensure new growth.Moving back upstream to higher-growth phases can be extremely difficult.Few tech companies,apart from Apple,have produced successful second acts in unrelated adjacencies,but Adobe,Microsoft,Nvidia,and others have renewed themselves with an upstream sh
41、ift by reigniting and modernizing the core.Late-growth companies such as Amazon have also successfully unleashed Engine 2 catalysts,generating a second wave of growth.Interest rate hikes have increased the cost of funding,so shareholders are watching returns on capital more carefully and rewarding c
42、ompanies that understand when to reinvest in the business,when to use capital more efficiently,and when to return capital.As businesses grow,their investors and valuations tend to reflect the maturity of their core business.Sometimes,smaller,less mature businesses trapped within the bigger company a
43、re not properly valued by the marketa fact thats often pointed out to boards by activist investors rather than management.Portfolio restructuring can unlock value and help match these businesses with the right investments.Companies should be proactive in shaping their investor base,targeting the typ
44、e of investors they want and tailoring communications to them.Amazon did this by signaling its intent to deprioritize profit in favor of continually reinvesting in the business.By clearly communicating these priorities,Amazon was able to set expectations and attract growth-oriented investors who sup
45、ported this reinvestment rather than conservative investors who would not have rewarded this strategy.To maximize value,strategies for growth,capital allocation,and investor relations should change over time as markets mature.Shaping the investor base and communicating clearly to set expectations ar
46、e essential to ensure investor support,whether growing the core business,searching for a new engine of growth,unlocking hidden assets,or improving capital efficiency.Understanding the roles among market maturity,investor expectations,and sources of TSR is essential to deliver shareholder value at ev
47、ery step of the journey.10Value EvolutionPreparing for Exit:A Buyers Market Is Coming for Tech AssetsA backlog of portfolio investments is likely to come to market when deal activity picks back up.By Brian Kmet,David Lipman,Christopher Perry,Jen Smith,and Colleen von Eckartsberg At a Glance Reduced
48、exits,extended hold times,and steady dry powder reserves point toward an incoming wave of tech asset exits that will create a crowded,competitive buyers market.Firms that have prepared to stand out will fare better than average,while others may struggle to sell mature assets in their portfolio.The v
49、olume of technology deals has slowed since mid-2022 for a number of reasons,including limited debt availability amid rising interest rates and declines in asset values that have left buyers unable to meet sellers asking prices.Successful deals have relied on greater equity contributions(with expecta
50、tions to refinance later),partial equity sales to fund growth,and a greater proportion of add-ons rather than standalone or platform assets.Overall,the pace of tech deals since the third quarter of 2022 continues to be slow,in line with the broader deal market.Exits are also downon average,about$20
51、billion per quarter in the first half of 2023 compared with$107 billion per quarter in the first half of 2021 and$75 billion per quarter in the first half of 2022.A growing backlog of deals,including more than$700 billion of tech assets purchased between 11Technology Report 2023Notes:YTD after year
52、stands for year-to-date;F after year stands for forecastSource:Dealogic(as of July 5,2023)Following buyout frenzy in 20202021,exits have slowedA backlog of deals will come to market when demandrecovers,including more than$700 billion of assets that traded between 2018 and 2021Q17365223Q157255825Q124
53、97985QQ1201302513Q0222023Global tech sector buyout exit value,in billions of US dollars(20182023YTD)Global tech sector buyout deal value,in billions of US dollars(20182023F)20320200222092023F158Likely to tradewithin thenext three yearsAnnualized
54、Figure 1:Tech deals have slowed over the past year,and a backlog of deals points to a coming buyers market2018 and 2021(see Figure 1),has led to longer hold times of tech portfolio companies.In 2023,nearly half of tech portfolio companies have been held for more than four years,and 15%have waited mo
55、re than six years.For the first time since 2012,more than 40%of tech portfolio companies are being held for more than four years(see Figure 2).This backlog of long-held portfolio assets is growing more quickly than the mountainous level of dry powder that is holding steady,which will create a buyers
56、 market when activity picks up(for more,read Bains“Stuck in Place:Private Equity Midyear Report 2023”).The backlog of long-held portfolio assets is growing more quickly than the mountainous level of dry powder that is holding steady,which will create a buyers market when activity picks up.12Technolo
57、gy Report 2023Note:Data aggregated by year of investment and time in portfolioSource:PitchBook(as of March 31,2023)Number of tech buyout-backed companies by time in portfolio(global)01234K200002%40%38%37%35%32%29%29%30%29%33%44%Percentageheld forlonger tha
58、nfour yearsLonger than six yearsFour to six yearsZero to three yearsTime in portfolioFigure 2:Nearly 45%of current tech portfolio companies have been held for longer than four yearsHow to add valueWith multiples trending downward and a competitive market on the horizon,investors should focus on addi
59、ng value to existing assets by growing earnings,redefining operational processes that may have been ignored for a while,raising efficiency with new tools,and seeking new areas of growth.Within software,we see several interdependent trends.Focus on margin improvement:As market valuations have decline
60、d,investors have shifted focus from growth at any cost to rewarding profitability.To expand margins,companies need to build scale,increase automation,and boost productivityall while managing the costs of growth.Operational metrics and benchmarks become more important as companies eliminate low-value
61、 work,automate less-complex tasks,and explore offshore opportunities.Address weaknesses in the go-to-market model:A slowing economy has renewed the focus on sales and marketing capabilities.By synchronizing sales,marketing,ops,and product functions,companies can design focused,repeatable sales plays
62、.Additionally,the emergence of product-led growth lets software companies engage customers with a self-service model(for more,read the Bain Brief“What It Really Takes to Develop Product-Led Growth”).13Technology Report 2023Sources:Bain coinvestment database;Bain Revenue System Average multiple on in
63、vested capital for investments by type of effortDeal countIndustry average2.0Diligence(pre-acquisition)2.5Post-acquisition3.0 19984Figure 3:Post-acquisition portfolio actions can deliver up to a three times multiple on invested capital Evaluate AI-driven disruption and efficiency improvements:The re
64、cent boom in consumer AI usage and business-to-business applications has led to a variety of new use cases,specifically in two key areas:R&D and enablement tools.Portfolio companies are evaluating the potential of generative AI,and they are focusing their R&D efforts on areas of growth while pulling
65、 back on distractions(for more,see the related chapter in this report,“AI Investors:Act Fast,Act Wisely”).Consider new growth vectors:Companies are developing the next phase of their growth strategies,which should include adjacencies,new regions,and buy and build through M&A.When to add valuePortfol
66、io activism pays off at any point in the investment process,from early-stage due diligence to preparing for exit.We have seen better-than-average returns,up to three times multiple on invested capital(MOIC),with post-acquisition activism(see Figure 3).The biggest returns come from engaging early,as
67、close to day one as possible,as well as engaging later,with plans to move from good to great before an exit.14Technology Report 2023 Engage early to strengthen capabilities.Setting strategy along with substantial support in capability-level execution can improve earnings and boost the exit price.One
68、 example of adding value soon after acquisition comes from a software-as-a-service company offering critical data and communications IT software for business clients that wanted to develop a plan for efficient growth.The company started by identifying prospects and developing a channel strategy that
69、 shifted focus to its highest-priority accounts.Among other actions,the company diagnosed sales coverage,accelerated the sales pipeline,and assessed discounting and price increase practices in order to maximize revenue.A detailed diagnostic across marketing channels helped it differentiate its messa
70、ge and target the right audience.Online,it ran structured A/B tests and improved the way it measured lead tracking and revenue projections,using benchmarks from OPEXEngine to compare performance across industries.Through these actions,the company identified opportunities to raise earnings by about$4
71、0 million,mostly through price increases and new market leads.Engage later with targeted actions.For companies already performing well,strategic investments that raise their game from good to great can deliver the greatest punchon average,a 50%increase in MOIC within a year of targeted engagement,mo
72、stly from the work leading to a successful exit at an attractive price.For example,one educational technology company,a leading provider of online test and assessment solutions,had already achieved significant organic growth based on the strength of its product.An inconsistent go-to-market approach
73、that included investing too much in accounts unlikely to deliver the necessary returns,however,prevented the company from achieving its full potential.The company improved its go-to-market strategy to capitalize on its strong product advantage and to gain market share before its sale.It assessed the
74、 competitive landscape and redefined M&A objectives to articulate a pathway to further growth.Pushing a faster execution helped the company get more out of its market traction and trajectory,which boosted equity value during the sales process.These actions took an already remarkable 7 times MOIC at
75、the start of the engagement and raised it to 14 times at exit.A new segmentation of customers offered clarity on service and buying behavior for 13,000 potential customers.The company crystallized its strategy,revisited ideas for adjacency and acquisitions,and improved competitive dynamics to create
76、 the right dialogue for exit.Within two years,the private equity owners completed a partial sale of the business,keeping a sizable chunk themselves given their newfound confidence in the business.Investors and portfolio managers that dont add significant value to their software assets will struggle
77、to sell among a growing slate of mature assets.Tech investors that have done the work necessary to improve earnings are more likely to succeed in the upcoming competitive buyers market with assets that stand out.15Value EvolutionAI Investors:Act Fast,Act WiselyLower barriers to experimentation open
78、the door to more disruption,but incumbents have advantages,too.By Sanjin Bicanic,David Lipman,Christopher Perry,Gene Rapoport,and Jue Wang At a Glance Investors enthusiasm for artificial intelligence is high,with AI and machine learning investments leading venture growth in the first half of 2023.Ac
79、ross industries,AI should increase the productivity of knowledge workers,and software companies already benefit from AI coding assistants.Investors should be assessing risks and opportunities across their portfolios,understanding where technology will create new advantages and where it might lead to
80、 share loss.Investments in technology flocked to early-stage generative AI companies in the first half of 2023,led by Microsofts$10 billion investment in OpenAI(see Figure 1).With the potential for sweeping changes to the tech sector,investors are rightly afraid of the ways that generative AI techno
81、logies can affect current and future tech assets.The excitement raises a number of possibly overwhelming choices for investors:How will AI affect our portfolio companies?Which business models will change,and what new opportunities will present themselves?How do we adjust diligence criteria for futur
82、e investments?Are there ways we should deploy generative AI to improve our own internal operations?16Technology Report 2023Note:Data as of July 12,2023;AI stands for artificial intelligence;ML stands for machine learningSource:Pitchbook Quarterly capital investments in venture and growth stage deals
83、,in billions of US dollars0100200$300BH1 2021AI/ML deal valueAll other deal value201H2 2021230H1 2022199H2 2022118H1 202311117%14%13%14%28%Figure 1:Artificial intelligence and machine learning solutions led venture and growth funding in the first half of 2023“Software is eating the world,but AI is g
84、oing to eat software.”Jensen Huang,CEO of NvidiaBroad impacts in softwareAs the nature of human-to-computer interactions evolves,customer expectations are growing.Generative AIpowered chat interfaces for applications and data simplify the user interface,increase the localization and personalization
85、of content,and open new routes to market.New products will emerge,many of which will automate and augment the work of people in specific roles across sectors(see Figure 2).Software developers,for example,will become more efficient as AI-coding assistants supplement their efforts.Workers in other rol
86、es,such as customer support,technical field services,and sales and marketing,could all be augmented by generative AI.Start-ups and other small companies with fewer resources may be able to deliver new products more rapidly when assisted by AI.17Technology Report 2023Sources:Aura;Bain analysisPercent
87、age of employees by functions level of automation/augmentation potential020406080100%Travel andleisureBusinessservicesFinancialservicesConsumerservicesTech,media,andtelecomEnergy andnaturalresourcesPharmaAviationIndustrialsRetailHigh potential for augmentation in roles in engineering,design,sales,ma
88、rketing,and PRAugmentation and automation potential,for example,in customer interactionsHigh potential for automation in roles including accounting,IT,and administrationLow impact on roles in HR,legal,and engineering,but not coding or leadershipNo impact or unclearFigure 2:Generative AI will have di
89、fferential impact,depending on the share of automatable and augmentative rolesExperimentation also becomes easier,and barriers to entry are lower given reduced development costs and democratized foundation models at enterprise companies disposal.With lower barriers to entry,cycle times come down,req
90、uiring incumbents to act quickly to capitalize on the advantages of differentiated data assets,entrenched customer access,and integration into user workflows.Generative AI,however,introduces both opportunities and risksfor example,new AI features,such as a ChatUX,make it easier for users to engage w
91、ith a companys product,but risks also emerge as users seek out other AI-enabled applications that might better address specific use cases,potentially reducing market share.Competitive landscape implicationsAlthough most investors agree that AI will have a significant effect on the tech sector,the ev
92、olution of the competitive landscape remains to be seen.Tools and enablers.Large language models(LLMs)and other foundation model providers are likely to consolidate,forming a few winners in each category.Open-source models are likely to be part of the evolving landscape,too.We also expect more conso
93、lidation for tools supporting generative AIincluding cloud providers,system integrators,and specialized semiconductorsas a result of the 18Technology Report 2023rising R&D investment required to maintain the pace of innovation.Many of todays largest tech companies will benefit,and leaders,including
94、cloud service providers,OpenAI,and Nvidia,are already seeing record-breaking growth.For other generative AI tools and enablers(including data and systems and services that facilitate the use of AI),the story is more nuanced.With few incumbent or leading providers,a large number of early-stage compan
95、ies are likely to arise to provide support for building LLM-based apps in categories such as data management,storage and process capabilities,and AI implementation services.These tools may eventually consolidate as larger platforms eventually provide these services in-house.Software applications.Bey
96、ond tools and enablers,there will be winners and losers among software applications innovating on new and existing use cases.With broad and inexpensive access to democratized foundation models,a flurry of early-stage players will likely develop innovative use cases using existing foundation models.T
97、hese will include vertical and horizontal applications for use cases that werent previously possible.With broad and inexpensive access to democratized foundation models,a flurry of early-stage players will likely develop innovative use cases using existing foundation models.Among incumbents,software
98、 companies that learn to deploy generative AI technology in relevant markets are likely to emerge as winners.Unlike the transition to cloud,the benefits of AI can often be realized without investing in major overhauls of company platforms.On balance,this favors incumbent software providers that have
99、 access to data,customer relationships,and a track record of execution.Customer access and customer data protects incumbents from disruption by new competitors and start-ups,but sustained market leadership will depend on how incumbents adopt generative AI to make their products and operations better
100、.A healthcare IT company encountered this situation as it evaluated opportunities to use generative AI across its product suite.Larger competitors were quickly harnessing generative AI,potentially putting the healthcare IT companys areas of differentiation at risk.Its position as a specialized provi
101、der could be threatened as customers begin to use generative AI tools and consider a broader set of vendors with features that would make their daily operations more productive.To counter those risks,the company set out to rapidly embed generative AI features that would enhance the customer experien
102、ce and further differentiate its products.19Technology Report 2023How can funds avoid disruption risks?Top funds arent waiting to see how generative AI changes this space.With shorter cycle times and lower barriers to entry,incumbent advantages will dissipate if they dont act now.In assessing whethe
103、r a market will face significant change from generative AI,investors must consider both disruption potential and structural barriers in the market.Does generative AI have the potential to replace or augment human effort,improve product quality,or reduce costs?What are the structural barriers to entr
104、y?Are there legal restrictions or sensitive data involved?In assessing the companys ability to capitalize on these opportunities,investors must consider whether they own proprietary data that could enrich generative AI applications.Is the companys pricing model set up to capture value from generativ
105、e AI,or will it face pricing pressure?Do you have the talent to executeand if not,where can you find it?What defensible moats,such as customer stickiness and brand awareness,can you lean on?By understanding the overall potential for change in markets from generative AI and the ability of assets to n
106、avigate that change,top funds are biasing toward action to capitalize on the potential of their incumbent software assets.20Technology Enters Its Post-Globalization EraHow tech companies are relocating manufacturing,R&D,and talent for resilience.By Karen Harris,Anne Hoecker,Christopher Schorling,Bal
107、aji Thirumalai,and Jue WangAt a Glance Tech companies are diversifying their supply chains,R&D locations,and talent pools in order to build resilience against global shocks.The semiconductor industry,for example,is responding to government incentives to make large,long-term investments in fabs in th
108、e US,Japan,and Germany.These changes will create new technology hubs and shift the competitive landscapepartly as a result,the hiring of engineers is accelerating in India and Europe.A decade from now,the global footprint of the technology value chain is likely to look very different,but companies s
109、hould be evaluating their options today.The shocks of the past few years have awakened tech companies to the risks of restrictive policies and export controls.Most are now moving rapidly to build resilience into their supply chains,primarily by expanding their geographic footprints and creating more
110、 flexibility within their talent pools.Value Evolution21Technology Report 2023Investing in resilience is costly and could take decades.But the good news is that resilience doesnt mean 100%risk aversionan approach that would be too expensive and inefficient.Practical resilience means diversifying the
111、 most critical aspects of your business while getting closer to end markets.Beyond supply chains:Disentangling talent and R&DThe initial impulse to be closer to end markets,diversify geographically,and respond to pressures from local government and regulatory bodies led companies first to disentangl
112、e their supply chains.Now,they are extending those efforts to R&D,talent,and innovation centers.The semiconductor industry is incentivizing the build-out of new fabs outside of historical locations,primarily targeted at newer chips at smaller node geometries,but new fabs take three to five years to
113、come online and produce chips in volume.Relocating manufacturing and component suppliers.Tech manufacturers that have their facilities concentrated in one region or that sell to markets on the opposite side of the world are boosting resilience and improving efficiency by diversifying their manufactu
114、ring footprint to new locations in Asia,Europe,and North America.It will take time for companies to move production,and they will also need to balance the risk of oversupply as new sites come online.The semiconductor industry is incentivizing the build-out of new fabs outside of historical locations
115、,primarily targeted at newer chips at smaller node geometries,but new fabs take three to five years to come online and produce chips in volume.Even downstream original equipment manufacturers and component suppliers that are moving production to new locations will require a few years to materialize
116、and deliver the same standards as original factories.Apple,for example,has started to diversify its manufacturing base outside of Chinaand India and Vietnam have been the biggest winners.India is expected to become a major manufacturer of iPhones soon.The percentage of iPhones produced in India grew
117、 from 2%in 2021 to about 5%in 2022,and this number is expected to grow to 25%.Several of Apples contractors and component suppliers,including Foxconn and Foxlink,are also planning to invest in new capacity in India to support this expansion(see Figure 1).Chipmaker TSMC has also started to expand its
118、 production capacity outside of Taiwan,with announcements to build multiple fabs in the US,Japan,and Germany,investing close to$65 billion.TSMC relies on local subsidies from recently passed legislation in those regions to cover 20%to 50%of the total cost.22Technology Report 2023In 2021,Lenovo doubl
119、ed its manufacturing footprint in Monterrey,Mexico,allowing the company to manufacture all data center products for North American customers within a Nafta country.This lets Lenovo manage its supply chain more effectively to better serve its North American customers.Foreign direct investment in Chin
120、a has gone down,with the notable exception of Tesla,which recently announced that it will open a second factory in Shanghai in 2024 to produce the Megapack battery.This is in addition to its Gigafactory that started production in 2019,which was built to provide local manufacturing for the companys s
121、econd biggest market outside of the US.This helps Tesla serve its large customer base in China,avoid tariffs,and source parts locally to reduce costs.Relocating R&D and talent.Restrictions moved faster than expected to impact talent and R&D.For example,in October 2022,the US Department of Commerces
122、Bureau of Industry and Security added new rules restricting the ability of US persons to support the development or production of integrated circuits at certain fabs in China.Tech companies are avoiding single points of failure and tapping markets with concentrations of specific skills(see Figure 2)
123、.They are strategically relocating business-critical functions and roles to areas with lower geopolitical risk.Approaches include duplicating high-value capabilities in several locations,documenting at-risk processes in case they need to be replicated in new locations,and hiring more employees in sa
124、fer geographies.Notes:F after year stands for forecast;supplier locations reported for fiscal year 2021 and fiscal year 2018Sources:Company reports and filings;Apple 2022 annual report;Apple 2019 corporate social responsibility reportApple final assembly volume by geographyApple supplier number of l
125、ocations020406080100%20212225F202225F202225F202225F020406080100%20182021China JapanUnited StatesTaiwanKoreaVietnamMalaysiaThailandSingaporePhilippinesGermanyIndiaOtheriPhonesAirPodsiPads andApple WatchMacBookChinaIndiaVietnamFigure 1:Apple is assembling more iPhones outside of China and diversifying
126、 its supplier base23Technology Report 2023Partly as a result of these strategies,India and Europe are seeing an acceleration in the hiring of engineers.India has become a thriving location for multinational companies to establish R&D centers in artificial intelligence(AI)because of its large talent
127、pool,which offers vast potential for growth.Among other examples,Fujitsu launched a new AI R&D center,Fujitsu Research of India Private Limited,in 2022 as part of its global innovation strategy.Implications for technology companiesGiven the high cost and time commitment to build resilience,some comp
128、anies are waiting on the sidelines as others take the first step.While its not essential to be the first mover,its certainly critical not to be the last to act.The key is to have a plan in place.Identify critical risk areas.Across the supply chain(manufacturing,component supplier,R&D,talent),compani
129、es should start by assessing which parts of the business are strategically most important and most susceptible to risks,along with identifying critical dependencies and where to build redundancy.Each business unit requires different types of resilience,with varying com-plexity and potential impact o
130、n earnings.Source:LinkedIn analysis includes headcount in engineering function in each location across computer hardware,software,networking,computer and network security,and semiconductors industriesPercentage of engineering headcount that joined the current company over the past two years(raw Link
131、edIn statistics)05101520%IndiaEuropeVietnamUSJapanTaiwanChinaTwo yearsOne yearLess than one yearEngineers tenures at current companyPartially caused by reduced LinkedInaccess in ChinaFigure 2:R&D relocation and nearshoring are accelerating the hiring of engineers in India and Europe24Technology Repo
132、rt 2023 Evaluate options.In assessing complexity,companies should consider locations where governments are making it easier to do business.Several countries and trading blocs are putting new rules in place and offering subsidies to encourage new,local technology supply chains that use government inv
133、estment dollars and tax credits.For example,recent regulatory initiatives in India are creating favorable grounds for manufacturing and talent relocation,and in the US and EU,regulations are paving the way for more semiconductor investments.Be ready to invest.Tech companies will have to coinvest and
134、 share know-how to support the re-balancing of supply chains,aiding in the broader development of new local supplier ecosystems.Specifically,this could include helping key suppliers get approval to build sites in new countries as well as providing the talent,engineering,and management expertise need
135、ed to build products to the necessary standards.Test the waters.Companies can start with smaller,less important products to assess markets,avoiding the risk of being a late mover.They can move to core products after gaining a foothold in new locations.A decade from now,the global footprint of the te
136、chnology value chain will look very different.Todays investments in resilience are likely to create new hubs and ecosystems while shifting the competitive landscape.Industries may fragment,and new local heroes will emerge.Now is the time to begin making early investments to ensure the resilience of
137、talent and supply chains.Youre Out of Time to Wait and See on AI .27How AI Is Recoding the Software Business Model .32Generative AI and Cybersecurity:Strengthening Both Defenses and Threats.37Taking the Hyperbole Out of the Metaverse .41The Untapped Value at the Intelligent Edge .47After the Chip Sh
138、ortage,Fears of a Capacity Glut Are Overblown .52Strategic Battlegrounds27Strategic BattlegroundsYoure Out of Time to Wait and See on AIArtificial intelligence is as much an economic revolution as a technological one.By David Crawford,Arjun Dutt,Eric Sheng,and Jue WangAt a Glance Its expensive to bu
139、ild the large foundation models that power generative AI,but once they exist,its relatively inexpensive to experiment with these models.The current generation of AI tools and models could help companies speed up 20%of worker tasks without a loss in quality.Engineering teams are already deploying AI
140、coding assistants,and 75%of executives told us that AI has already met or exceeded their expectations.Beyond taking current work and making it better,faster,and cheaper,the real value of AI is likely to come from new uses that werent possible before.Few technology revolutions have rolled out this qu
141、ickly.Five years ago,machine intelligence lagged far behind humans.But over the past year,general purpose foundation models have demonstrated capabilities rivaling a persons ability to perceive,understand,communicate,create,reason,and employ tools.The public release of ChatGPT late in 28Technology R
142、eport 2023Note:Floating point operations show the computing power necessary to train the models and thus is also a proxy for cost Source:Bain&Company20020202022510260.010.GPT(2018)BERT(2018)AlphaGo(2017)GPT2(2019)DALL-E(2021)Amazon AlexaTM(2022)
143、Google Chinchilla(2022)Bloomberg-GPT(2023)Google GShard(2020)GPT3(2020)StableDiffusion(2022)GPT4(2023,directional)PaLM 540B(2022)Model training(floating point operations)Model size(billions of parameters)Threshold for modeling the content of the InternetFigure 1:Foundation models have grown in size
144、and cost over the past six years2022 made these models accessible to a large base that had no previous experience with artificial intelligence(AI).Two months after its public release,ChatGPT had gained more than 100 million users,an adoption rate five times faster than that of previous champ TikTok.
145、Of course,AI is an overnight success story decades in the making.We are still in the early days of one of the most important technology inflection points in history,but in this fast-moving environment,adopting a wait-and-see posture is as good as being left behind.Most significantly for the enterpri
146、se,foundation models and AI represent much more than a technology revolution;theres also an economic revolution underway.While the cost of building the foundation models is high and has increased as the models have grown larger(see Figure 1),their existence creates a relatively low bar for entry for
147、 companies to experiment with ways to use foundation models and generative AI in their business.Until recently,AI work(and talent)has been concentrated among the big cloud service providers.Now,with the dramatic reduction in the economics of AI experimentation and application,its spilling over into
148、a broader spectrum of industries and capabilities,where innovation is happening quickly and already creating new value.29Technology Report 2023How are tech leaders assessing their opportunities in foundation models and generative AI?Bains recent research reveals how technology executives are approac
149、hing this opportunity,where theyre investing and experimenting,and what the most likely disruptive plays could be in the months ahead.Assessing potentialThe most recent foundation models improved dramatically over previous generations,demonstrating better language abilities and scoring higher on tes
150、ts.This dramatic change in economics and performance to understand,reason,generate,and act on unstructured data is revolutionizing knowledge work in the enterprise.Bain research suggests that access to a large language model and AI tools could help companies speed up 20%of worker tasks without a los
151、s in quality.Companies are already deploying foundation models and generative AI tools to reduce costs,increase productivity,and enhance qualitycheaper,faster,and better.(For more,see the Bain infographic“The Era of Enterprise AI Is Here.”)of engineering teams say that they are deployingcoding assis
152、tants for software development57%Enterprises are seeing early results and productivity gains in applications,including sales and marketing,IT ticket resolution,HR talent acquisition and performance management,software development,and customer contact centers and help desks.For example,in Bains recen
153、t research with software companies in the process of evaluating and rolling out generative AIbased coding assistants,75%of executives said that AI met or exceeded their goals,76%said it was speeding time to market,and 63%said it improved the quality of code.(For more,see the Bain infographic“AI Codi
154、ng Assistants Are Already Living Up to Expectations.”)System integrators and IT services lead the pack in most disrupted industries and fastest adopters for activities such as writing code,supporting field services,and managing IT(see Figure 2).Beyond“cheaper,faster,and better,”foundation models and
155、 generative AI are already creating new value in cases that werent possible before.Rapid upskilling:At one company,new customer service agents were able to reach productivity levels that normally take six months to attain in only two months because they were supported by generative AI models that su
156、ggested responses for them based on summaries of a vast database of answers.30Technology Report 2023Source:Bain AI Survey,2023(n=571)Adoption pace by industryPercentage of respondents currently adopting or evaluating at least one of the top six foundation model use casesAuto and transport60%Media an
157、d entertainment63%Energy/natural resources68%Retail69%Financial services73%Consumer goods76%Healthcare/life sciences79%Advanced manufacturing79%Telecommuncations82%Technology85%Marketing/advertising86%System integrators90%AllOne to two use casesThree to four use casesFive or more use cases78%Figure
158、2:IT system integrators lead the way in AI adoption,closely followed by companies in marketing,advertising,and technology Hyper-personalization and A/B testing at marginal cost:For example,Khan Academys AI-powered guide Khanmigo helps students understand complex concepts by adapting to explain these
159、 concepts at various levels,depending on the students degree of understanding.Redefining the customer experience:Generative AI could fundamentally change how customers interact with software and data.For example,New Relic launched Grok,a feature aimed at allowing users responsible for monitoring app
160、lications and infrastructure to interrogate deep system telemetry data through natural language,aiming to dramatically simplify a typically sophisticated,complex user experience.Reimagining the business definition:Generative AI capabilities can expand the value of transactional apps by accessing mor
161、e data about the user and relevant external data.For example,Carrefour introduced a shopping assistant called Hopla that does more than simply generate shopping lists and integrate with its website;it suggests new items and menu ideas based on budgets and diets.Core product innovation:Companies are
162、using generative AI to radically reduce product development times and add new features to products.For example,pharmaceutical companies are using generative AI to speed up development times.Insilico Medicine,for example,deployed 31Technology Report 2023the technology to help identify a molecule that
163、 a drug compound could target and bond with,helping it predict the outcome of clinical trials.The company said using generative AI allowed it to accomplish these tasks in one-third the time and for one-tenth the cost.In game development,Roblox recently announced the development of a platform to enab
164、le users to create sophisticated 3D content via voice,text,or touch prompts by using generative AI,dramatically simplifying and democratizing the creator experience.Building momentumFor senior executives and boards of tech companies,the arrival of AI at scale presents enormous opportunities.Most are
165、 closely focused on understanding how AI can add value,but its also important to tend to a range of technical and process issues that can support rapid deployment of AI solutions.Among the technical aspects to consider:Are privacy and security mechanisms in place to ensure security for data and cust
166、omers,as well as compliance with regulations?Do we have a solid build vs.buy strategy to ensure that we build only where it differentiates us while maintaining the right balance of vendor choices to avoid lock-in?Do we have the right machine learning operations(ML Ops)and engineering processes to te
167、st and maintain the quality and resilience of complex,non-deterministic models,as well as the quality of input data?And do we have a scalable architecture for reusable components and models?Do we have access to enough computing resources,including a wide spectrum of power and performance design poin
168、ts from the cloud to the edge?Process questions include:Will our rapid experimentation in sandbox environments deliver quick wins that demonstrate potential value?Do we have a roadmap for scaling and change management that will help us capture the full potential value beyond the first experiments?Ar
169、e we prepared to redesign business processes and redeploy labor as material innovations require?The rapid uptake of AI over the past year and the many ways that it contributes to productivity and creativity have generated a broad wave of enthusiasm across the entire tech sector and beyond.Products a
170、nd services seem likely to evolve quickly over the next six months,particularly since the establishment of large foundation models means that the cost for experimentation is relatively low.Waiting to see what competitors will do is tantamount to yielding the field.How will your company create new va
171、lue with AI today?32Strategic BattlegroundsHow AI Is Recoding the Software Business ModelNew features could accelerate growth,but changes in customer processes and job roles could disrupt established use cases and pricing models.By Simon Heap,Christopher Schorling,and Ted SheltonAt a Glance Easy acc
172、ess to large language models has unleashed a torrent of experimentation in the software industry:Nearly 9 out of 10 developers are already using artificial intelligence(AI)in their products.Software vendors expect a step change in functionality,and those that move too slowly could lose competitive a
173、dvantage.The rise of AI also poses some risks for software companiesfor instance,AI tools could make it easier for customers to develop their own software instead of buying it.If AI reduces the number of roles in some job categories,software vendors that price on a per-seat basis may need to adjust
174、their business models.Over the past six months,nearly every software company has announced plans to apply generative artificial intelligence(AI)within their business.The ease of access to large language model(LLM)application programming interfaces(APIs)has made it relatively easy to demonstrate new
175、AI-powered products.Bains recent survey found that 89%of software companies are building AI into their products to differentiate them15 percentage points higher than other sectors.33Technology Report 2023Source:2023 Bain AI Survey(overall n=571,enterprise software companies n=99)New AI-based product
176、s and servicesNew AI-powered featuresNatural language user interface Percentage of respondentsTop-linegrowthImprovedcustomerretentionImprovedcustomerretentionImprovedcustomerretentionHeadcountefficiency/productivityHeadcountefficiency/productivityHeadcountefficiency/productivityFastertime tomarketFa
177、stertime tomarketFastertime tomarketOpexreductionOpexreductionOpexreduction82%472083Percentage of respondentsTop-linegrowthTop-linegrowth77%611444Percentage of respondents55%4038139Figure 1:Software companies see benefits from new artificial intelligence products,features,and interfacesInnovation is
178、 happening quickly,and were still in early days.Three out of four software companies surveyed believe that early movers will have a sustained advantage that will not level off.Software leaders expect the technology to generate significant opportunities to increase top-line growth and customer retent
179、ion(see Figure 1).All that can be certain is those that fail to start the journey will be left behind.Opportunities and risksSoftware companies also need to address how the adoption of generative AI by their customers and competitors can threaten their core business.Some software categories may face
180、 an existential threat as AI fundamentally changes the capabilities and economics of what can be done in areas such as low-or no-code application development and customer experience management.Most application categories will experience a step change in functionality and user experience,so vendors s
181、hould prioritize their AI roadmaps to deliver their own capabilities where they will count the most.As customers introduce AI into their own processes,job roles will change.This could lead to a dramatic fall in the number of end users,which would hurt segments that rely on seat-based pricing,such as
182、 service desks,engineering development,marketing and project management,and others 34Technology Report 2023(see Figure 2).Seat-license software companies can respond by ensuring that the product still adds value as customers reengineer their processes;they can also adopt an alternative,consumption-b
183、ased pricing model(for more,read the Bain Brief,“Is Consumption-Based Pricing Right for Your Software?”).Developing an understanding of how customers are adopting AI and reengineering their processes can help identify other risks and opportunities.Nearly 40%of the companies we surveyed said that the
184、y are either evaluating or adopting foundation models or generative AI into their work.Most activity is in software development,knowledge work,and content creation aimed at helping companies synthesize data,create content,or assist in reasoning and planning.Rather than simply automating discrete rol
185、es(for example,an AI-powered chatbot),were seeing a complete reimagining of workflows and job roles.For example,AI could help a product manager create marketing content,thereby reducing the need for some downstream roles in product marketing and content authoring.In the past,vertical software vendor
186、s focused on particular industries and relied on process expertise to maintain a competitive advantage.That edge could become harder to maintain as horizontal software vendors are able to use LLMs to interpret industry knowledge and develop industry features at low cost,using their scale and scope t
187、o beat niche players.Sources:Bain&Company;IDC;company websitesPercentage of revenue that top 10 independent software vendors accrued through seat-based pricing(business-to-business software)Content workflow andmanagement applications89%Enterprise resourcemanagement applications83Customer relationshi
188、pmanagement applications82Analytics and businessintelligence software81Application developmentsoftware77Security software74Figure 2:Generative artificial intelligence will have a greater effect on software categories that rely on seat-based pricing models35Technology Report 2023AI and product strate
189、gyAs software companies think about how AI will change their business,they should consider several important questions.How will generative AI change the customers businesses,and what does that mean for the software product and business model?Automating process steps is simple,but how will customers
190、radically reimagine their workflows?Will our AI-powered solutions be differentiated by a proprietary LLM?Or will we differentiate through the integration with other systems and data?How will improvements in R&D productivity affect the speed of innovation?Will LLMs and generative AI enable a new leve
191、l of customization for users,reversing the trend toward standardization?Will users still interact with our application,or will it be intermediated by a new chat-based user interface?These choices will help identify where differentiation will matter most,how to construct the user interface,and how to
192、 package and price products.LLM investment barriers to entry are dropping rapidly,and the open-source community continues to innovate in kind.Options to partner or build are expanding,and some customers might have preferences for or against models hosted by particular cloud service providers.Some us
193、e cases are better served with narrower models rather than general foundation models such as GPT4.For example,Intuit is rolling out a generative AI operating system on which its next generation of applications will be built.The OS would be based on a narrow LLM targeting tax,accounting,cash,and pers
194、onal finance tasks.Although companies such as Salesforce and ServiceNow use LLMs from third-party vendors,they are also building their own LLMs.As expected,propensity to build and train an LLM increases with company size(see Figure 3).Many customer concerns around data protection and access,personal
195、ly identifiable information,audit trails,prompt grounding with proprietary data,and integration with other machine learning and automation technologies are served in platform layers,beyond the LLM.Its here that many software companies can differentiate themselves,leveraging established positions in
196、customer architectures.For example,Salesforce is investing in its LLM Gateway architecture while accessing GPT4 and other LLMs through APIs,where they also have MuleSoft API management capabilities.Generative AI will transform the way users interact with software.They no longer will have to interact
197、 through traditional user interfaces,and they can now do so with natural language via 36Technology Report 2023Source:Bain Tech Talent Survey,2023(n=75)Percentage of respondentsMy organization will pre-traina large language model andbuild its own solutionMy organization will use anexisting open-sourc
198、e largelanguage modelMy organization will use theapplication programminginterface of a commerciallarge language model020406080100%$50 million to$99 millionin revenue6$100 million to$499 millionin revenue20$500 million to$1 billionin revenue17More than$1 billionin revenue32BuildBuyBuyFigure 3:Larger
199、companies are more likely than smaller ones to build their own large language models,but even they are more likely to buya chatbot.Software makers decisions about whether this is a simpler user interface for existing applications,a new user portal for all process interactions,or an established gener
200、al purpose text interface(such as Slack or Teams)can shift to where the control point sits.Software makers also have to decide how to package and price new AI functionality.Whether its a separate product,an add-on,or just embedded as an enhancement of the existing product,independent software vendor
201、s could monetize it with a price increase,or they might choose to treat this as continued innovation that pays for itself by retaining customers.But beware the gross margin impact of running computing-intensive features.Generative AI is already having a huge impact on the software industry,offering
202、both opportunities and threats to the established order.Software vendors must embrace the opportunities and not shy away from reinventing themselves to remain relevant and capitalize on this new technological development.Doing so requires understanding how customers will reengineer themselves around
203、 existing applications,and that will help software makers decide how to architect and differentiate new AI solutions.37Strategic BattlegroundsGenerative AI and Cybersecurity:Strengthening Both Defenses and ThreatsBreakthroughs in technologies built on large language models will accelerate the arms r
204、ace between hackers and companies.By Syed Ali and Frank FordAt a Glance Generative artificial intelligence(AI)should strengthen cybersecurity,particularly in threat identification,although its unlikely to lead to full automation anytime soon.Bad actors are also exploring generative AIs potential to
205、aid cyberattacks through innovations such as self-evolving malware.Through a range of moves today,both buyers and providers of cybersecurity services can take advantage of the new technology while remaining protected.Only months after its public breakthrough,generative AI has shown the potential to
206、transform cyber-security products and operations.After the launch of ChatGPT and other products powered by large language models(LLMs),the cybersecurity industry is planning for generative AI to become a key tool.And thats despite the launch challenge generative AI faces in cybersecuritynamely,the s
207、ensitive and siloed nature of security data,which makes it hard to get high-quality,comprehensive datasets to train and update an LLM model.38Technology Report 2023Notes:Percentages rounded;analysis is of cybersecurity companies that are using generative AI to enhance solutions Source:Bain&CompanyCu
208、rrentFull potential Little impactModerate impactHigh impactImpact of generative AI by incident response stagePreparationIdentificationContainmentEradicationRecoveryLessons learnedUsed by 20%of analyzed cybersecurity companiesUsed by 100%of analyzed cybersecurity companiesUsed by 65%of analyzed cyber
209、security companiesUsed by 55%of analyzed cybersecurity companiesUsed by 50%of analyzed cybersecurity companiesUsed by 40%of analyzed cybersecurity companiesGenerative AI should streamline cybersecurity analyst trainingThreat detection and hunting will become more dynamic and automatedContainment of
210、lower-level threats could be further automated,but full automation unlikely in next decade,if everRecommended responses to lower-level threats could be further automated,but full automation unlikely in next decade,if everRecommen-dations and best-practice instructions could be further automated,but
211、full automation unlikely in next decade,if everIncident response reportswill be much improved,but organizational and process changes will still need humansFigure 1:Threat identification holds the most potential for generative AI to improve cybersecurityand thats where industry adoption has been stro
212、ngest so farSo far,threat identification is the hot spot.When we analyzed cybersecurity companies that are using generative AI,we found that all were using it at the identification stage of the SANS Institutes well-known incident response frameworkthe biggest uptake in any of the six SANS stages(pre
213、paration,identification,containment,eradication,recovery,and lessons learned).That fits our assessment that threat identification holds the greatest potential for generative AI to improve cybersecurity(see Figure 1).Generative AI is already helping analysts spot an attack faster,then better assess i
214、ts scale and potential impact.For instance,it can help analysts more efficiently filter incident alerts,rejecting false positives.Generative AIs ability to detect and hunt threats will only get more dynamic and automated.For the containment,eradication,and recovery stages of the SANS framework,adopt
215、ion rates vary from about one-half to two-thirds of the cybersecurity companies we analyzed,with containment most advanced.In these stages,generative AI is already narrowing knowledge gaps by providing analysts with remedy and recovery instructions based on proven tactics from past incidents.While t
216、here will be more gains through automation of containment,eradication,and recovery plans,full automation is unlikely over the next 5 to 10 years,if at all.The longer-term impact of generative AI in these areas is likely to be moderate and will likely always need some human supervision.39Technology R
217、eport 2023Sources:Rapid7;Bain&CompanyNumber of dark web mentions of generative AI01,0002,0003,0004,000Q3 2021Q4 2021Q1 2022Q2 2022Q3 2022Q4 2022Q1 2023ChatGPTFigure 2:The use of generative AI for nefarious purposes has become an increasingly popular topic on the dark web after the launch of ChatGPTG
218、enerative AI is also being used in the lessons-learned stage,where it can automate the creation of incident response reports,improving internal communication.Crucially,the reports can be reincorporated into the model,improving defenses.For example,Googles Security AI Workbench,powered by the Sec-PaL
219、M 2 LLM,converts raw data from recent attacks into machine-readable and human-readable threat intelligence that can accelerate responses(under human supervision).But while the quality of generative AIpowered incident response reports should keep improving,human involvement is still likely to remain
220、necessary.A double-edged swordOf course,generative AI can also be used as a cyberattackers tool,giving them similar capabilities as defenders.For example,less experienced attackers can use it to create more enticing emails or more realistic deepfake videos,recordings,and images to send to phishing t
221、argets.Generative AI also allows bad actors to easily rewrite a known attack code to be just different enough to avoid detection.Generative AI has certainly become a trending topic for malicious actors.Mentions of generative AI on the dark web proliferated in 2023(see Figure 2).Its common to see hac
222、kers boasting that theyre 40Technology Report 2023using ChatGPT.One hacker posted that he was able to use generative AI to recreate malware strains from research publications,such as a Python-based stealer that can search and retrieve common file types(.docx,PDF,images)across a system.The threat fro
223、m bad actors will only increase as they use generative AI to standardize and update their tactics,techniques,and procedures.Generative AIassisted dangers include strains of malware that self-evolve,creating variations to attack a specific target with a unique technique,payload,and polymorphic code t
224、hats undetectable by existing security measures.Only the most agile cybersecurity operations will stay ahead.Actions to take nowCorporate leaders should:understand that generative AI wont rid cybersecurity of its operational and technical complexities;make generative AI and cybersecurity a recurring
225、 agenda item for board and C-suite meetings;and avoid a narrow focus on controls or certain riskscybersecurity demands a holistic approach.Chief information officers/chief information security officers should:get security operations(SecOps)leaders to validate generative AI output,particularly threat
226、-detection algorithms updated by generative AI;train new and junior SecOps employees to hunt threats with and without generative AI to avoid dependence;and where possible,avoid relying on a single vendor or generative AI model across the cybersecurity stack.Cybersecurity companies should:hire the ri
227、ght mix of talent to bring generative AI capabilities into their products;and guard against generative AIcreated false information(hallucinations)and external tampering with generative AI algorithms and models that might create backdoor vulnerabilities.Generative AI will rapidly advance,and its esse
228、ntial that all stakeholders from cybersecurity providers to enterprises continuously update their specialist knowledge and strategy to take advantageand stay protected.41Strategic BattlegroundsTaking the Hyperbole Out of the MetaverseThe metaverse isnt dead,but its time to take a more sober view of
229、its future.By Chris Johnson,David Crawford,Nihar Naik,and Neha VermaAt a Glance The metaverse is here,with immersive gaming platforms boasting hundreds of millions of monthly active users,and emerging metaverse technologies deployed across industries.Bain projects that the metaverse could reach a ma
230、rket size of$700 billion to$900 billion by 2030 and that the major winners will likely be determined in that time frame.While its unclear how the competitive landscape will shift,five tech battlegrounds will shape the metaverse:virtual experiences,content-creation tools,app stores and operating syst
231、ems,devices,and computing and infrastructure.The debate over what the metaverse will ultimately become misses the more salient point:The metaverse is already here.People imagine a future straight from the sci-fi novel-turned-movie Ready Player One,in which we would spend most of our time living in c
232、ompletely immersive virtual worlds.But when we break it down,the metaverse is a collection of immersive,collaborative,and interactive environments that span digital and physical worlds and enable us to create,exchange,socialize,and learn.And that part of the vision already exists.42Technology Report
233、 2023Popular gaming platforms todayFortnite,Roblox,and Minecraftare becoming more interactive,enable users to create avatars and worlds within them,and deepen the level of immersion players experience by delivering increasingly photorealistic content.Theyre also doing this on a massive scale.Each of
234、 the platforms mentioned has around 200 million monthly active users,on par with the respective number of subscribers of Amazon Prime and Disney+.In the business world,a suite of emerging metaverse technologies that leverage virtual reality(VR)and augmented reality(AR)are already providing value.The
235、se include applications in areas of high risk(e.g.,AR-assisted spinal surgery)and high expense(e.g.,jet engine manufacturing)in which interacting with digital models can help improve care or save money.While its challenging to predict exactly how the metaverse will look decades from now,its clear th
236、at the journey is well underway and theres a real and growing economic opportunity.We project the metaverse could reach a market size of$700 billion to$900 billion by 2030.Furthermore,the next 5 to 10 years will likely determine the metaverses major winners(more on this below).A multiyear tech boom
237、and free-flowing venture capital fueled a lot of excitement about the metaverse over the past few years.Now,as executives map their strategies in a very different economic environment,we think framing the discussion through the following lenses will help determine where to focus their metaverse inve
238、stments.The metaverse will take years to reach full scaleThe metaverse will likely remain in the first phase of developmentwhat we call the“seed”stage of the platform development journeyfor at least another 5 to 10 years(see Figure 1).Reaching scale requires bringing compelling,breakthrough use case
239、s to market in a format that enhances user experiences.It took more than 10 years for smartphones and gaming consoles to each exit the seed stage and begin scaling up,while it took almost 20 years for the Internet and personal computers,respectively.Nevertheless,it wouldnt be surprising if the metav
240、erse grew faster than other influential technologies,given how the pace of change has accelerated.And if a company waits for the metaverse to reach the scale phase before it enters the market,its likely already too late.Big market winners often emerge during the seed stage:think Microsoft and Apple
241、in PCs,Nintendo in video games,or Apples iPhone in smartphones and iPad in tablets.Today,a mix of metaverse strategies exist,from companies with a vertically integrated approach that spans multiple segments of the metaverse technology stack to those with a horizontal strategy that focuses on a singl
242、e layer of the stack.It remains to be seen which will prove to be most effective as the market develops,and the shape of the ecosystem(vertical vs.horizontal)is likely to evolve over time.Meta has established an early lead among those focused on vertical integration,combining leading hardware,app st
243、ores,and virtual content in pursuit of widespread user adoption and engagement.The companys vertical approach allows for tight coupling of hardware and content and,therefore,strong curation of the user experience.43Technology Report 2023Device installed base/Internet users(B)Years from technologys l
244、aunchSeedScaleExpandNotes:Game consoles user base declined as gaming PCs market grew;smartphones defined as mobile phones that have a high-level operating system with third-party native applications and connectivity;smartphone installed base decreased 3%during the Covid-19 pandemic,as second or thir
245、d phones(e.g.,company phones)were redundant with decreased mobility Sources:Euromonitor;IMF reports;Gartner;Bain analysis54325302520151050InternetPCs20SmartphonesMetaverseGame consolesInternetPCsSeedScaleExpandFinding initial product-market fitAccelerating adoptionReaching adjacent market
246、s and revenue streamsFigure 1:The metaverse may take 5 to 10 years to begin scaling,based on the evolution of analo-gous technologiesAnother notable company with a vertical strategy is Apple.It entered the market in June with the announcement of a mixed-reality headset that combines AR and VR,uses A
247、pple-designed chips and an Apple-designed operating system,and connects with Apples existing app ecosystem.Companies that have adopted a more horizontal approach include Nvidia and Epic Games.Rather than build end-to-end technology stacks,they supply silicon and content-creation tools,respectively,t
248、hat are used by many companies in the metaverse ecosystem.The metaverse will be pluralTodays consumer and enterprise applications will become increasingly immersive and collaborative,moving closer to the most aspirational vision of the metaverse by unlocking richer experiences and better functionali
249、ty(see Figure 2).And while its unclear how the competitive landscape will shift and potentially converge,its unlikely to become one big platform dubbed“the metaverse.”Although open standards initiatives have attracted the attention of major tech companies,including Microsoft and Meta,an industrywide
250、 shift to a truly interoperable metaverse likely wont happen anytime soon.Similar to todays Internet,private companies have strong incentives to maintain“walled gardens”and the companies proprietary datasets within them so that they can effectively monetize and recoup the investments required to dev
251、elop metaverse technologies.44Technology Report 2023Osso VRZwiftYouTubeInstagramDiscordHorizon WorkroomsGoogle WorkspaceAVRTGenshin ImpactPokmonGOVSCOPhotoshop(Adobe)VRChatHorizon WorldsRec RoomOffice 365RobloxWoWGustoText-basedgames Immersion levelInteractivityMoreClassVRLessAsynchronousSynchronous
252、Virtual spaceNotes:Immersion level determined by immersive nature of content(70%of weight)and physical and digital world blending(30%of weight);interactivity determined by user-to-user interactiveness(70%of weight),with the remaining 30%split among the ability to create user-generated content and th
253、e presence of digital economies for consumer applications;enterprise applications werent considered for the digital economies criteria given different monetization models;Google Workspace includes Google Meet(video conferencing)and has more monthly active users relative to its enterprise counterpart
254、s due to greater consumer adoption;Office 365 is bundled and includes Microsoft Word,PowerPoint,Excel,OneDrive,and Teams;text-based games include Choice of Robots,Reigns,Sorcery!,The Dreamhold,and others Sources:Company websites;press releases;public company financial earnings calls;Bain analysisMon
255、thly active users1M10M100M100M500M500M1B1M10M1B+Primarily enterprisePrimarily consumerMedium(Adobe)MatterportSMStextingFacetimeTwitchZoomMinecraftTrelloFortniteNiche use casesSmaller immersive platforms likely to remain focused on specific industry verticals1Social destinationsMetaverse-first worlds
256、 trying to achieve scale through a broad set of social,gaming,and productivity experiences2Creative gamingMature gaming worlds expanding to social use cases partly through user-generated content3Virtual workplacesEnterprise companies racing to add new collaboration features43124Figure 2:Todays consu
257、mer and enterprise apps already exhibit metaverse attributes and are evolving quickly45Technology Report 2023Rather,platforms with large user bases today may take steps to become increasingly immersive and engaging,while smaller,metaverse-like worlds will try to attract bigger user bases.The size of
258、 these walled gardens will depend on their ability to tap into network effects and spur user-generated content,kicking off a flywheel effect that accelerates the platforms growth.These efforts wont preclude open standards initiatives from making progress,but walled gardens will likely remain more pr
259、evalent for the foreseeable future.Five battlegrounds are shaping the metaverseMetaverse profit pools are likely to accrue across the technology stack at key control points(see Figure 3).In short strokes,heres where the most consequential competitive dynamics are taking place.Virtual experiences(abo
260、ut 65%of metaverse projected market size in 2030):Although gaming is currently the leading consumer metaverse application,immersive fitness and entertainment could also be compelling in the medium term.On the enterprise side,innovative use cases are emerging,primarily in collaboration and productivi
261、ty but also in digital marketing,employee training,education,and healthcare.Source:Bain&CompanyUser-facingVirtual experiencesContent-creation toolsApp stores and operating systemsDevicesComputing and infrastructureImmersive virtual environments in which users(both consumer and enterprise)can interac
262、t,consume content,play games,collaborate,learn,and run simulationsTools to facilitate the creation of realistic virtual environments via 3D graphics rendering,audio,and other features that mimic the physical worldCritical access point for users to explore,find,and download apps and worlds in the met
263、averse,and for companies to curate the metaverse experience Enable users to enter the metaverse,through traditional devices such as PCs,phones,or tablets,or in 3D via augmented or virtual reality devicesSemiconductors,servers,and networking technology enabling computing,rendering,and connectivity fo
264、r real-time,multiplayer participation across devices0102030405Technologyback endFigure 3:In the tech industry,there are five key metaverse battlegrounds46Technology Report 2023 Content-creation tools(about 5%of metaverse market size in 2030):Theres a growing field of software tools that provide the
265、building blocks,editing platforms,and interfaces for creating metaverse worlds and experiences.Many leading platforms offer simple drag-and-drop world-building features that are typically free and intended to cultivate user engagement,retention,and monetization.Generative AI will also transform the
266、way content is created by making it easier to create avatars,in-world assets,and interactions with virtual characters.Meanwhile,developer tools continue to advance,both for consumers and enterprises.App stores and operating systems(about 10%of metaverse market size in 2030):We expect the metaverse t
267、o follow the smartphone model,with app stores serving as the primary access points for users to explore apps and worlds in the metaverse.The app store role will be even more crucial during the metaverses seed stage,helping provide users with curated,high-quality experiences to keep them engaged with
268、 the platform and headset they use to access the metaverse;this need is amplified by the immersive nature of VR content.Devices(about 10%of metaverse market size in 2030):Significant technological barriers must be overcome before the arrival of comfortable,lightweight,standalone devices that allow f
269、or truly immersive experiences.To achieve mass adoption,metaverse content will need to work across all types of devicesincluding,for the foreseeable future,personal computers,gaming consoles,and smartphones.These device manufacturers may see their businesses benefit as some consumers opt for their h
270、igher-end product lines to unlock more immersive,collaborative experiences.Computing and infrastructure(about 10%of metaverse market size in 2030):Demands on computing power will continue to grow significantly.That will pressure hardware companies to develop higher-performing chips,servers,and netwo
271、rking technologies to render high-quality graphics and reduce latency.Improvements are already underway,including 5G network rollouts and advances in system-on-chip performance.No one knows exactly what the metaverses final destination will be,but all signs indicate its going to be a journey well wo
272、rth taking.While executives cant ignore the metaverse,neither are we suggesting that every company pivot its entire strategy to focus on this sector.Its more about making targeted adjustments to existing road-maps to ensure that the company moves in the same direction as its ecosystem of customers,p
273、artners,and competitors.No one knows exactly what the metaverses final destination will be,but all signs indicate its going to be a journey well worth taking.47Strategic BattlegroundsThe Untapped Value at the Intelligent EdgeThe need to process data close to its source is creating a new computing ar
274、chitecture and a growing market.By David Crawford,David Deming,and Bill Radzevych At a Glance The market for intelligent edge computing is greater than previously understood,with as much as$127 billion in spending on embedded silicon forecast through 2027.This increase reflects the rise in the intel
275、ligent edge paradigm,which puts computing resources close to data sources to reduce latency,power usage,and bandwidth costs.Succeeding at the intelligent edge requires strong capabilities to integrate software,digital hardware,and analog hardware since customers give higher satisfaction ratings to f
276、ully integrated solutions.Edge computingthat is,moving processing from central data centers to regional and local computing clusters closer to the sourcerepresents a major shift in technology.Because of constraints on power and latency,many applications benefit from data processing at the sourcefor
277、example,close to sensors(see Figure 1).For these applications,a new architecture is emerging:intelligent edge devices.Sending vast amounts of data to regional or local cloud hubs for processing is inefficient at best and has the potential to overwhelm existing transport and computing architectures.F
278、urthermore,much of the data generated is noise,and filtering and processing it right at the source can cut down 48Technology Report 2023Sources:Gartner IoT Semiconductors Forecast(worldwide third-quarter 2022);Bain&CompanyPrimary drivers of intelligent edge computingMarket for embedded silicon in in
279、telligent edge applicationsOtherInfotainment(voice,audio,camera)Automotive(lighting,body control,climate control)Smart watchesOtherHead-mounted displays(augmented reality/virtual reality)HearablesMilitary electronicsOtherAutomation edge device/embedded AIAdvanced driver assistance system moduleAuton
280、omous driving system(levels zero to three)LatencyDecision making moving closer to the sensor improves latency$127 billionPower/form factorComputing-enabled sensors optimize power consumption and reduce form factorData volume reductionComputing at the edge sends out only useful data,significantly red
281、ucing the volume of data transmissionFigure 1:Power constraints,latency requirements,and limits on connectivity shift computing toward an intelligent edge paradigmdramatically on transport and processing requirements.Many applications also have a very low tolerance for latency because they are unabl
282、e to wait for transfer times or delays,requiring resolution in real time.Even traditional edge processing nodes,including local data centers and on-premise gateways,are not efficient enough for the real-time feedback loops these modern devices require.Intelligent edge devices represent a paradigm sh
283、ift as they perform most data inference and processing directly at the device itself and offer distinct advantages in terms of latency,power consumption,and use of bandwidth.Noise-canceling headphones are a good example:They analyze sound signals in the surrounding environment and counter with oppos
284、ite frequencies locally,reducing background noise without the need to transmit data to the cloud.Video surveillance is another example:Instead of sending a raw video pixel by pixel to a central processor,which would require tremendous bandwidth and central processing power,an intelligent edge camera
285、 can perform a first-order structuring of the datafor instance,looking for humansand send this structured data to a central hub for decision making.Not all devices will use this architecture,of course.For example,large medical datasets that search for genetic abnormalities require comprehensive proc
286、essing and inference,and they need access to the massively parallel computational power offered in cloud architectures.Additionally,many smart building management systems take advantage of a central hub in the building to make decisions based on sensor data coming from all over the building.49Techno
287、logy Report 2023Percentage of developers that rank criteria among the top four most importantNote:Net Promoter Score?is a service mark of Bain&Company,Inc.,NICE Systems,Inc.,and Fred ReichheldSource:Bain Intelligent Edge Survey(n=140)59%53%51%41%33%Software operating systemSecurityOptimal power perf
288、ormanceSoftware software developer kitIntegrated full-stack solution(hardware and software)Average Net Promoter Score?for intelligent edge solutionsIntegrated hardware and software solutionsVendor-driven partner ecosystems 12522Figure 2:Developers prioritize integrated full-stack solutionsBut for th
289、e devices at the intelligent edge,the market opportunity may be larger than previously understood.Bains analysis forecasts that by 2027,$127 billion of the$277 billion total market for embedded silicon could come from the intelligent edge.Its a huge opportunity,but most ecosystem participants arent
290、fully ready to address this market.Given how rapidly this space is evolving,companies need to react now to serve this market effectively.Develop new capabilitiesTapping the intelligent edge opportunity requires new technical capabilities and business models.First,companies will need to further devel
291、op their ability to integrate software with digital and analog hardware to deliver fully integrated solutions.Bain research finds that when customers choose solutions,integration of hardware and software is the most highly ranked purchase criteria and that customers award a higher Net Promoter Score
292、sm for fully integrated products,though even here there is room for improvement(see Figure 2).Given their starting points,this will mean finding the right partnerships or acquisitions while also building up internal capabilities.But our research is clear:Delivering an integrated,end-to-end solution
293、will produce the product most likely to win in these markets.50Technology Report 2023These companies will also need to strengthen their abilities to assess and determine which solutions merit full integration and where it makes more sense to deliver parts of solutions,leaving integration to the cust
294、omer or third-party integrators.As markets mature,customers may develop a preference for open solutions that they can more easily tailor to their needs or that fit better into their software ecosystem.Understanding the determining factor among digital,analog,or a combination can help shape the best
295、offer for a given application and customer.Developing a better understanding of customers and their commercial needs will also be key.Its likely that as the market matures,customers will want more control over the elements of a solution.Understanding the determining factor among digital,analog,or a
296、combination can help shape the best offer for a given application and customer.Capture opportunitiesGiven the importance of integration and the nascent nature of the solutions in this space,traditional lines of competition are blurring in intelligent edge computing.Most vendors do not provide full-s
297、tack solutions,so to win in this market,companies have to find partners across the edge solution stack.For example,application-specific solutions that integrate digital and analog components bring together microcontroller and analog players.Gateway players such as Nvidia and Qualcomm are investing i
298、n integrated computing and software solutions as well as developer-friendly tools(for example,Nvidia Jetson and Qualcomm Drive Data Platform).We see many participants in the market starting from their position of strength,pushing into adjacent segments,and blurring traditional lines of competition(s
299、ee Figure 3).Companies in different tech sectors have different avenues to success in the intelligent edge market.Digital semiconductor:Digital and connectivity players should continue to work toward the integration of analog components,strengthening the integrated software stack and building purpos
300、e-specific applications.Analog players:Anchor in analog strengths while leveraging the ecosystem and partnerships to build full-stack solutions and software capabilities.Software:Meet developers in their ecosystem of choice,and embrace growing industry standards and open-source offerings.51Technolog
301、y Report 2023 Original equipment manufacturers:Focus on strategic issues and holistic use cases,and let vendors absorb the hardware and software complexity.Hyperscalers:Recognize where cloud-tethered solutions have limitations,and partner with hardware vendorsfor instance,as NXP partnered with Amazo
302、n Web Services.Intelligent edge devices represent a new computing paradigm that will enable a new class of devices that react in real time to the world around them,without needing the computing and connectivity requirements of the cloud or local data center processing.But success will also require a
303、 shift in capabilities and new linkages within the ecosystem.Vendors in this space should assess their readiness for this shift.Notes:GPU stands for graphics processing unit;AIPU stands for artificial intelligence processing unit;ML stands for machine learning Source:Bain analysisEdgeIntelligentedge
304、Sensor ConnectivityGeneral purpose processingAccelerated processing(GPU/AIPU)Application development platforms(including AI/ML)Application softwareCloudGatewayOther digital players(e.g.,Intel,AMD)Digital and connectivity players(e.g.,Nvidia,Qualcomm)Traditional digital players(e.g.,NXP,Renesas,Infin
305、eon,STMicro)Traditional analog players(e.g.,Analog Devices,Texas Instruments)Cloud service providers(e.g.,Amazon Web Services,Microsoft Azure,Google Cloud Platform)Figure 3:The intelligent edge requires more integrated solutions,pushing players to expand their capabilities52At a Glance The semicondu
306、ctor industrys post-pandemic rebound boosted capacity to the extent that some foresee an oversupply.But cyclicality in the sector is normal,and its actually been leveling out over time as the end markets for chips have broadened and suppliers have consolidated.Even so,supply-and-demand balance could
307、 remain bumpy in three areas that customers should monitor:bleeding-edge,leading-edge,and super-lagging chips.Companies can guard against the risks of imbalance by designing products for resilience and flexibility,taking advantage of moments of oversupply,and developing chip-specific strategies for
308、mission-critical components.Many manufacturers that depend on the semiconductor supply chain are breathing easier now.As post-pandemic demand for PCs,smartphones,and consumer devices eased,capacity grew in the semi market.In fact,weve gone from a chip shortage in 20212022 to underutilization of capa
309、city in some parts of the semi value chain.Some say this is a harbinger of a massive capacity glut,implying we no longer need to worry about securing our semiconductor supply chain.We disagree.The industry and its customers are adept at balancing long-term capacity and supply.By David Crawford,Peter
310、 Hanbury,Anne Hoecker,and Michael SchallehnAfter the Chip Shortage,Fears of a Capacity Glut Are OverblownStrategic Battlegrounds53Technology Report 2023Note:F after year stands for forecastSource:Gartner Worldwide Capital Spending Forecast,second quarter 2023Year-over-year percentage changes for sel
311、ected semiconductor ecosystem metrics,global GDP included for scale(19962027F)50050973200520072009200023F 2025F 2027F100Global GDPSemiconductor market revenueWafer fab equipment spendingFigure 1:Cyclicality has moderated in the semiconductor marketCyclical
312、ity and brief periods of underutilization are normal features of the semiconductor business cycle.But cyclicality has actually dampened over the past couple of decades for three main reasons(see Figure 1).First,the law of large numbers suggests that the broad set of end markets,each with different d
313、emand cycles,helps counter-balance demand.Second,although memory is the most cyclical part of the semi market,it now represents a smaller portion of total capex.Finally,fewer suppliers of logic and memory result in a less irrational supply.Cyclicality is likely to continue,but structural or long-ter
314、m overcapacity is unlikely for several reasons.One is the underlying secular growth of the industry combined with the pragmatic nature of semi-conductor manufacturers when it comes to building out capacity.Bleeding-edge fab shells(7-nanometer nodes and below)are not cheap,costing about$2 billion and
315、 taking more than two years to build.Even greater expense lies in the equipment,which can run up to$9 billion and take up to 18 months to qualify and ramp up.When downturns reduce expectations to fill capacity,manufacturers often hit pause before adding machinery to prevent periods of overcapacity.T
316、his creates large swings of 54Technology Report 2023cyclicality in the semiconductor equipment market as projects are started and then paused abruptly.As the market for electronics returns to its historic growth,demand for semiconductors grows,quickly consuming any overcapacity,which is why we rarel
317、y see semiconductor downcycles lasting longer than two years.As the market for electronics returns to its historic growth,demand for semiconductors grows,quickly consuming any overcapacity,which is why we rarely see semiconductor downcycles lasting longer than two years.Another factor that absorbs o
318、vercapacity is that chips can be designed on a range of technologies,and products can be adapted to take advantage of available supply,thereby rebalancing demand.This redesign process takes 12 to 18 months.So,although it didnt help alleviate the recent shortage,it does help prevent structural or lon
319、g-term overcapacity.Similarly,large government subsidies dont usually change this calculus.Although the US and EU are offering about$100 billion in subsidies through 2030,this is only a fraction of the$1 trillion in capex that the semiconductor industry plans to invest.Government subsidies might inf
320、luence the location of fabs built,but they are unlikely to lead to structural overcapacity.Since the market is growing,overbuilt capacity is generally consumed by market growth within a year or two,as we have seen in previous cycles.Gauging supply,demand,and pockets of overcapacityAlthough cyclicali
321、ty is leveling out over time,pockets of overcapacity still occur for short periods(one or two years)and for certain technology nodes or regions.The big question here is whether the semiconductor market remains a single global demand pool or splinters into competing blocs.A key indicator will be whet
322、her governments start to mandate where some chips are manufactured(domestically or via a trade partner).This would accelerate the splintering of semiconductor ecosystems and have a greater effect than subsidies on the global market for semiconductors.We also see specific technologies in which the cu
323、rrent plan to build out supply could lead to challengesfor example,the bleeding-edge foundry capacity being added by TSMC,Intel,and Samsung in the US,or the older technologies largely being built out in China and Taiwan.55Technology Report 2023Given these dynamics,we expect that the supply-demand ba
324、lance will be bumpy over time by node and by region,especially in three key areas.Bleeding-edge foundry:If Samsung and Intel take some of TSMCs customers or if government subsidies broaden the base of semi suppliers(such as the new Rapidus group in Japan),this would make capacity expansion decisions
325、 significantly more complex and prone to miscalculation,with overinvestment in boom times and underinvestment in busts.Leading-edge technologies:In the Asia-Pacific region,many foundries(TSMC,UMC,SMIC,Samsung,GloFo)produce mid-tier mature nodes(28-nanometer to 90-nanometer nodes).The market for 28-n
326、anometer nodes,in particular,is expected to see capacity grow at 12%annually over the next few years vs.5%to 7%for 45-nanometer and 90-nanometer technologies.If significant capacity comes online at the same time,we could see overcapacity for these older nodes for a period of time.Super-lagging edge
327、fabs will remain a challenge:Theres a long tail of low-end chips,usually produced on 6-and 8-inch wafers,that serves a wide range of applications from industrial and defense to automotive and consumer devices.Similar to how low-cost labor makes producing a cost-competitive T-shirt in the US challeng
328、ing,the economics of depreciated fabs producing low-end chips make it harder to get new lagging-edge fabs built in nearshore locations.About 60%of new super-lagging edge foundry capacity is expected to be built in China.Companies that rely on semiconductors should continue to design for resilience a
329、nd traceability.What does all this mean for industries that rely on semiconductor chips?Customers still need to pay attention to the semiconductor supply chain.Deepen your semiconductor capabilities.The semiconductor market remains complex,with many moving pieces.Companies that rely on semiconductor
330、s should continue to design for resilience and traceability.Improving understanding and engagement with the semiconductor market is critical.Leverage moments of oversupply.Close monitoring of fab build-outs and demand can help identify pockets of oversupply in which players should secure capacity an
331、d take advantage of favorable pricing opportunities.56Technology Report 2023478Notes:F after year stands for forecast;consumer electronics includes tablets;accelerator cards have been included in data center segment,which includes logic,analog,discrete,opto,and sensors and excludes memorySource:Gart
332、ner Semiconductor Forecast Database,second quarter 2023Worldwide semiconductor sales,excluding memory(in billions of US dollars)0200400$600B20220224562024F2026FData centerNetworkingAutomotiveIndustrialPCConsumerelectronicsSmartphoneOther545Figure 2:Applications beyond classic IT are growi
333、ng faster Develop chip-specific strategies,part 1.For example,in low-end chips,an area where multiple suppliers can be qualified,its important to monitor the entire market.Unexpected demand spikes in emerging areas(for example,electric vehicles)could create shortages of specialty chips.Semi vendors have historically focused on the classic IT market of computers and other tech devices.But automotiv