The Deflation of Software

Generative AI has the potential to deflate the software industry.

The proliferation of generative-AI-powered software development is widely expected to catalyze digital transformation across industry sectors and unlock broad economic growth. If generative AI continues to make software development more efficient and accessible, businesses could increasingly replace today’s packaged software with generative-AI-powered solutions. This could reduce the economic power and influence of many of today’s software companies — which are primarily US-based¹ — relative to other companies and industries, and lead to economic drags in countries that are exporters of software and development services. Such upheaval could drive M&A activity within software and the broader technology industry, and the business of software — the preeminence of the software industry and the organizing principles of software companies — could change.

Companies across all industries should take note because such a “deflation of software” could be a precursor to the obviation of larger categories of knowledge work. Software development has been quickly disrupted by generative AI because the enabling datasets, namely source code repositories, have been easily identified and utilized. Over time, similar phenomena may play out in other areas of knowledge work, with even larger geopolitical and economically redistributive implications.

What does it mean for the software industry to deflate?

In a 2011 essay, venture capitalist Marc Andreessen coined the expression "software is eating the world," observing that software companies were capturing ever-larger shares of value and growth versus other sectors of the economy.² Thirteen years later, the prescience of Andreessen’s statement is well known. Over the five years ending in 2021, the software industry grew 14% per year to $630 billion in revenue, far outpacing the growth of broader IT and the global economy and capturing an increasing proportion of global profit and value creation.³ Companies in every industry spent increasing amounts on software – today, software is perhaps the largest capital spending category in the U.S.⁴ – catalyzed by the emergence of cloud-based deployment models and punctuated by enterprises’ response to the COVID pandemic. The exclamation mark was 2021’s IPO market: of the 111 technology IPOs Goldman Sachs tracked, half were for software companies, totaling over $29bn in capital raised and $319bn in new public company value.⁵

Today, generative AI is expected to have similar transformational effects and to capture increasing amounts of spend across industries, but this may not be to the benefit of the software industry. Software companies now face a multitude of pressures — economic and technological — that make their further growth much more challenging.

• Generative Al is unlocking new levels of efficiency and cost-effectiveness in software development. Traditional software development is labor-intensive and time-consuming, involving numerous stages of human involvement from ideation to deployment. Generative AI is already automating significant aspects of coding, testing, and maintenance, with Microsoft GitHub Co-PiIot being the most prominent example of this phenomenon. GitHub claims that 46% of new code is already developed by AI.⁶ As generative software development becomes more pervasive, the ability of individuals, teams and enterprises to build their own custom software improves, and the pricing power of packaged software providers decreases. As I said in my 2023 “The Insight” video, enterprises “might not have to buy so much software.”⁷

•  IT customers are reducing budgets for software products and seeking vendor consolidation. Today’s belt-tightening among the customers of software appears to be both cyclical and secular — cyclical, since COVID pulled demand forward while today’s higher-for-longer interest rates militate reduced spending, and secular, since customers are shifting budget to generative AI, the bulk of spend for which accrues to hyperscalers, semiconductors and systems.⁸ If investment in generative AI grows at very high rates while the growth of the overall IT industry remains low, cannibalization may occur. Arguably this would be less of a challenge for the largest software companies, who have entrenched customer relationships and datasets that can power generative AI functionality (and who in some cases are hyperscalers themselves), versus the lower end of the industry.

• Relatively simple software applications could be obviated by generative-AI-powered solutions. Technologies commonly utilized by SaaS companies of the cloud era included browser-based interfaces, relational databases and public cloud infrastructure. Many SaaS applications add value by using these technologies to streamline workflows that would otherwise be manual. The infrastructure stack for generative AI is very different – much more reliant on natural interfaces, vector databases and GPU infrastructure. Over time, businesses may develop generative-AI-native solutions for various needs, potentially replicating the functionality of SaaS products, substituting for existing software packages and eventually altogether replacing certain tasks and roles.

The degree to which generative-AI-powered software will replace existing software packages is a critical assumption. Incumbency in a software business can include at least four elements: “data gravity,” meaning the data within a given system or application are typically challenging to extract and move for use elsewhere; user engagement, meaning the familiarity and attention of an existing human user base is difficult to replicate; customer relationships, meaning access to the relevant stakeholders at a customer organization; and business logic, meaning having taken the effort to specify and implement a system for a given value proposition. Generative AI is already disruptive to business logic incumbency in the sense that it has already become much more efficient to write code. But the degree of disruption to the other elements of incumbency could greatly affect generative AI’s long-term effect on the software industry. For example, if data gravity within SaaS applications persists, then SaaS businesses may persist for a long time – even if their products become increasingly “wrapped” with generative interfaces and the businesses look increasingly like data warehouses.

We are already witnessing investors turn broadly away from software, expressing the view that the growth of the sector may be in secular deacceleration even as certain semiconductor and systems manufacturers accelerate. Software investors are increasingly focused on generative AI implications and the role each software company will play in the new AI era. Nvidia recently (albeit temporarily) became not just the world’s most valuable semiconductor company but the world’s most valuable company overall.⁹ Almost contemporaneously, the world’s largest SaaS company, Salesforce, missed revenue growth expectations and traded down 20% in a single day; it now trades at just ~70% of its all-time high, reached just 5 months ago.¹⁰

What is the macroeconomic and geopolitical significance of a deflating software industry?

The software industry today is highly globalized and US-centric, built on a foundation of free trade and interoperability, supported by the economic might of large software companies. A small number of multinational companies (e.g. Microsoft, Oracle and Salesforce based in the U.S. and SAP in Germany) represent the majority of software revenue worldwide and an effective global consensus for enterprise IT.¹¹ The U.S. enjoys a strong position in global policymaking, and is often the de facto decisionmaker, in matters of technology and Internet governance. Meanwhile, certain other countries, such as India, the Philippines, Brazil and Poland, achieve economic benefits as low-cost development centers and exporters of software and related services.¹²

A deflated software industry could have a number of economic and geopolitical effects:

• Benefits for challengers of the U.S.-led international order. We have already seen efforts by China to disrupt the software status quo and reshape the industry according to their own interests. Perhaps most notable has been the China Standards 2035 strategy, which aimed to set global standards for software development, AI and other emerging technologies.¹³ Equally significant are bans from time to time on various foreign software products. Other governments are similarly concerned about values and biases embedded within various technologies owing to the U.S.’ preeminent role. If generative software development or generative AI technologies make it easier to replace packaged and often-imported software products, the leverage of the U.S. in matters of technology could be diminished.

It should be noted that the geopolitical implications of generative AI are subject to the critical variable of infrastructure, namely the availability of Nvidia GPUs. Today, utilization of generative AI is essentially limited by the availability of these devices, in stark contrast to cloud computing, for which the critical hardware (CPUs and memory) is commoditized and abundant globally. Accordingly, the U.S. has instituted export controls, given the possibility that GPUs remain an effective control point for generative AI technology. It is not certain that Nvidia chips will maintain their differentiation over the long term, or that generative AI will remain so reliant on GPUs, or that the U.S. will be able to enforce export restrictions effectively; but if so, this could mitigate some of the threats to U.S. software hegemony. 

• Profits of the software industry may be redistributed to customers, and beneficial investment by software companies may be reduced. As the customers of today’s software companies replace existing products with generative-AI-powered solutions, software company profits may come under pressure. Query whether the redistributed profits will effectively be reinvested in economically beneficial ways. The potential knock-on effects of this upheaval could be massive: in the US, as of 2020, software companies represented ~27% of domestic business R&D spend, and nearly 1 in 10 jobs attaches to the software economy.¹⁴ Obviated workers will need to be reallocated. One needn’t look farther than San Francisco to witness the societal effects of a reduced workforce. 

• Countries that lack large software developer populations but have financial resources to invest in generative AI technologies may “catch up” in digital transformation and achieve corresponding economic growth. This could include resource-rich countries in the Middle East and advanced Asian economies.

• Demand for offshore development services may shrink. This could result in reallocation of profit away from countries that are exporters of software and IT services, most notably the US and India, with corresponding implications for economic growth. Private equity has historically played a role in offshoring the development activities of software companies to achieve financial returns; in the future, sponsors may similarly be catalysts for generative software development.

What will deflation mean for software companies?

The cloud era featured astoundingly fertile conditions for software company creation and proliferation. The shared economics of hyperscale meant that cloud-native software companies could be asset-light and their niches could be small. The 101 software companies that went public in the U.S. from 2016 – 2021 featured a median ~$200mm in trailing revenue at the time of IPO.¹⁵ Their greatest expenditures were typically on sales and software development, and they were typically lossmaking.¹⁶ The ability of a “team” to quickly bring products and features to market is what mattered most.

In today’s changed conditions, we are no longer seeing a proliferation of software businesses. Just four software companies have gone public since 2021.¹⁷ Investors are now looking for companies oriented toward the generative AI era, in which economies of scale in datasets and customer relationships are critical, access to capital and GPUs is a differentiator and headcount is heavily scrutinized. In some ways, generative software development is arguably favorable to emerging software companies, because it reduces the resources necessary to develop any piece of software.  But at the same time, it levels the development playing field and thereby makes it harder for product teams to differentiate. Moreover, the dynamics around data gravity, customer relationships and user engagement generally favor large incumbents.

In the past, technological progress has spurred the rearchitecting of software companies. One example is what cloud did in the field of speech recognition, particularly in the case of Nuance Communications. Founded in 1992 as Visioneer, Nuance aimed to commercialize a proprietary set of speech recognition algorithms, which over time resulted in a multi-vertical portfolio of businesses: speech recognition for healthcare, for customer service, for automobiles, for productivity, and so on.¹⁸ By the mid-2010s, when speech recognition had essentially become commoditized by the public clouds, a company organized around speech recognition technology no longer made business sense. After that, the value of Nuance lay in the deep customer relationships and the highly specialized and verticalized datasets and integrations it had built up over three decades. So over several years, Nuance management skillfully divested and spun off various pieces of the company until Microsoft bought what remained (primarily the healthcare business) for $20bn in 2021.¹⁹

Similar rearchitecting may occur among today’s software companies as they attempt to navigate this more challenging operating environment. Consolidation is the most probable type of move that companies will make, as it can help achieve the benefits of scale while creating an opportunity to right-size GTM and R&D organizations. We are seeing increasing interest in this trend develop among both public and VC-backed software companies. We are also seeing consolidation continue in IT services, consistent with the direction of that industry for the last decade.

In parallel with the consolidation of subscale software companies, we may see increased acquisition interest in companies whose assets become more strategic in the context of generative AI. The clearest examples of this are high-quality datasets and/or models that can be utilized to map real-world business problems to generative solutions. GitHub, as a repository of open-source code, and Nuance, in the field of healthcare, have proved to be such assets for Microsoft; and we have already seen similar “dataset acquisitions” in the field of legal research.²⁰ One possible area of interest is IT services organizations: their software development know-how, if it is accessible, structured and high-quality, could be valuable in the context of generative software in the same way that GitHub has been valuable for accelerating software development via Co-Pilot.

In Conclusion

Software ate the world in the decade leading up to COVID; but in the decade ahead, the software industry will likely have to get leaner. We are unlikely to witness rampant software company proliferation for the foreseeable future, and the relative economic significance of small extant software companies is likely to wane. The world of software development will become “flatter,” potentially eroding a source of U.S. soft power, and attention will turn to other technological control points and sources of value in IT companies. We may see consolidation and reductions in investment across the software industry, which could result in the need to reallocate large numbers of workers and portend even larger trends across broader information work in general. Onward.

 

Disclaimer
This article has been prepared by Goldman Sachs Global Institute and is not a product of Goldman Sachs Global Investment Research. This article for your information only and should not be copied, distributed, published, or reproduced, in whole or in part. This article does not purport to contain a comprehensive overview of Goldman Sachs’ products and offerings. The views and opinions expressed here are those of the author and may differ from the views and opinions of other departments or divisions of Goldman Sachs and its affiliates. This article should not be used as a basis for trading in the securities or loans of any companies named herein or for any other investment decision and does not constitute an offer to sell the securities or loans of the companies named herein or a solicitation of proxies or votes. Goldman Sachs is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient. Certain information contained here may constitute “forward-looking statements” and there is no guarantee that these results will be achieved. Goldman Sachs has no obligation to provide any updates or changes to the information herein. Neither Goldman Sachs nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this article and any liability therefore (including in respect of direct, indirect, or consequential loss or damage) is expressly disclaimed. 

¹ 9 of the 10 largest software companies covered by GS Global Investment Research, as measured by market capitalization, are domiciled in the United States.

²Andreessen Horowitz, “Why Software is Eating the World,” Marc Andreessen (20 August 2011). https://a16z.com/why-software-is-eating-the-world/

³ Gartner, “Forecast Enterprise Infrastructure Software” and “Forecast Enterprise Application Software” (2024).

⁴ Goldman Sachs Global Investment Research, “Cyclical or Structural: Don't Count Software Out of Gen-AI,” Kash Rangan (3 June 2024). https://publishing.gs.com/content/research/en/reports/2024/06/03/aa51c538-31f3-4d1f-a7e0-365cc79024c0.html

⁵ Company public filings, Goldman Sachs Investment Banking calculations.

⁶ GitHub, “GitHub Copilot now has a better AI model and new capabilities,” (2023). https://github.blog/ai-and-ml/github-copilot/github-copilot-now-has-a-better-ai-model-and-new-capabilities/

⁷ Goldman Sachs, “Embracing Generative AI to Unlock Value,” (13 April 2023). https://www.goldmansachs.com/what-we-do/investment-banking/pages/embracing-generative-ai-to-unlock-value.html

⁸ Goldman Sachs Global Macro Research, “Top of Mind: The Post-Pandemic Future of Work,” (29 July 2021). https://www.goldmansachs.com/insights/top-of-mind/the-post-pandemic-future-of-work

⁹  FactSet, public trading data, Goldman Sachs Investment Banking calculation.

¹⁰ Ibid.

¹¹ Gartner, “Market Share Analysis: Enterprise Software, Worldwide,” (2022). https://www.gartner.com/en/documents/4801931

¹² Dreamix IT Consulting, “Software Outsourcing Market Size by Country,” (2024). https://dreamix.eu/insights/2024-software-outsourcing-market-size-by-country/

¹³ Center for Security & Emerging Technology, “Translation: The Chinese Communist Party Central Committee and the State Council Publish the ‘National Standardization Development Outline,’” (2021). https://cset.georgetown.edu/wp-content/uploads/t0406_standardization_outline_EN.pdf

¹⁴ Software.org, BSA Foundation, “Software: Supporting US Through COVID”, (2021). https://software.org/wp-content/uploads/2021SoftwareJobs.pdf

¹⁵ Company public filings, Goldman Sachs Investment Banking calculations.

¹⁶ Ibid.

¹⁷ Ibid.

¹⁸ Forbes, “There’s Nothing Nuanced About Microsoft’s Plans for Voice Recognition Technology,” (13 April 2021). https://www.forbes.com/sites/enriquedans/2021/04/13/theres-nothing-nuanced-about-microsofts-plans-for-voice-recognition-technology/

¹⁹ Microsoft, “Microsoft completes acquisition of Nuance, ushering in new era of outcomes-based AI,“ (2022). https://news.microsoft.com/2022/03/04/microsoft-completes-acquisition-of-nuance-ushering-in-new-era-of-outcomes-based-ai/

²⁰ GitHub, “GitHub Copilot Trust Center.” https://resources.github.com/copilot-trust-center/