Ketan Ahuja writes that recent government wins against Big Tech demonstrate the prudence of proactively stopping consolidation in the emerging AI startup ecosystem to prevent harm to innovation and consumers. OpenAI’s recent acquisition of Windsurf raises these competition concerns. The antitrust agencies should block it.
In their early years, Amazon, Google, and Facebook hoovered up startups in the emerging ecosystems for e-commerce, ad tech and social networking. They swallowed the best companies that threatened to trammel their growth and, through this strategy, built their modern dominance. Big Tech’s gain was society’s loss: forgone innovation and lower-quality services for consumers and business users.
Now, the antitrust agencies are seeking to unwind these acquisitions. The Department of Justice recently won its case against Google for monopolizing the digital advertising technology market and is seeking to force Google to divest its strategic acquisitions of DoubleClick and AdMeld that allowed it to establish dominance over online advertising. Similarly, the Federal Trade Commission’s lawsuit against Meta for monopolizing personal social media through its acquisitions of Instagram and WhatsApp has just begun its trial phase.
Hindsight is twenty-twenty. The digital advertising and social media markets were nascent, and regulators did not understand how these acquisitions would harm consumers and innovation. But now that the lesson is learned, regulators must more fiercely preserve competition in today’s rapidly developing artificial technology sector. This begins with blocking OpenAI’s acquisition of Windsurf.
The emerging AI ecosystem
Two years ago, experts anticipated that generative artificial intelligence models would become an extremely concentrated market. The most critical part of these AI models are the large language models (LLMs), which digest large amounts of data to recombine and reproduce information upon users’ requests. LLMs require so much data and financial resources to train that it makes economic sense to only have a few of them. Incumbent Big Tech companies, like Google and Microsoft, and leading AI startups, such as OpenAI, that took the lead to produce powerful models were expected to quickly carve up the market and establish their dominance.
Fortunately, now the LLM layer of the generative AI tech stack looks set to be less of a source of market power than anticipated. There are now several models available, and high-quality fully or partially open-source models such as DeepSeek and Llama limit the power of commercial model providers. This means that the AI model layer might become relatively commoditized.
A lot of the value in the AI ecosystem is now anticipated to be in the application layer: the software tools that harness and direct LLMs to perform useful functions such as coding and graphic design. Tyna Eloundou et al. estimate that LLMs alone will impact up to 15% of American workers’ tasks, but industry-specific software tools that harness LLMs could impact between 47% and 56% of workers. Venture capitalists are therefore pouring huge sums into companies that contextualize LLMs in this way.
Rather than through LLMs, market power in the AI ecosystem will instead cohere around products that contextualize LLMs for use in specific circumstances and build relationships with sets of users. As these products become embedded in user workflows and company systems in a path-dependent way, they will become very difficult for competitors to dislodge. This is why OpenAI’s acquisition of Windsurf, an otherwise small AI startup but one which commands a significant share of the fast-growing AI coding market, should concern regulators.
Generative AI as used in coding
Today, generative AI’s killer use cases are coding agents. While LLMs can be used for many different purposes, they are currently used most intensively to write and edit code. These models have gotten so good at coding that they can do so semi-autonomously at the direction of engineers who communicate with them largely in plain English, and sometimes through spoken conversations. For many simple tasks, humans often no longer need to touch code directly.
This has spawned a host of AI coding agent applications to facilitate coding requests. The frontrunners are Cursor and Windsurf. These companies put a software wrapper around a base LLM (such as GPT, Claude, Gemini, or Deepseek), which harnesses and steers the base model into an autonomous coding agent. These companies are among the fastest-growing of all time. The category leader Cursor, which OpenAI initially sought to acquire, is worth $10 billion, even though it was founded only in 2023. Windsurf is in second place, valued at $3 billion after its acquisition by OpenAI. Anthropic has a capable coding agent that runs within the terminal (Claude Code).OpenAI has also been developing a similar agent (Codex), raising the question of why it needs to acquire Windsurf in the first place. It is hard to estimate the market size of this fast-growing category, but it has billions in funding and looks set to significantly reshape software development, which employs nearly two million people in the United States.
Agencies should ban OpenAI’s acquisition of Windsurf
There are three main reasons why OpenAI’s acquisition of Windsurf raises competitive concerns and should be blocked. First, OpenAI may tie use of Windsurf to its LLM, GPT, potentially blocking a distribution channel for other LLMs (such as Anthropic’s Claude) into the coding agent market, and possibly throttling other coding agents that rely on OpenAI’s GPT LLMs. Second, it would remove a potential competitor from the AI coding market, especially given OpenAI’s experiments around building its own tool. Third, maintaining a greater number of significant independent AI companies, with habituated user bases and the capabilities to innovate in AI, contributes in the long term to a stronger AI innovation ecosystem.
Currently, coding agents allow users to select which base model they want to use. This makes it easy for users to switch between GPT, Claude, Gemini, or other LLMs. This interoperability supports competition among model providers in the AI tech stack. It also means that coding agents do not directly compete for the most part with LLM providers, although Anthropic and OpenAI do have their own terminal-based coding agents.
If OpenAI acquires Windsurf (and a significant portion of the coding agent market), it will have the incentive to break up interoperability and tie the use of Windsurf to the use of GPT. This would cut off a critical distribution channel for other LLM developers. In particular, Windsurf has been an important distribution channel for Anthropic’s Claude model. OpenAI will also have the incentive to pressure other coding agents to exclusively use GPT. This would build GPT’s market power in the LLM market and potentially the coding agent market. This closely mirrors the theory of harm in the Google Search and Google Ad Tech cases, in which Google used its financial power to dominate distribution channels and forced advertisers to use its digital ad auction market if they wanted to access its online ad space.
There is a good argument that antitrust enforcers should discourage the powerful LLM providers from acquiring any company in the AI application markets. Interoperability between platforms and applications has been a critical remedy sought in Big Tech antitrust cases, and it is a core pillar of the Digital Markets Act in Europe. Allowing LLMs to vertically integrate through the AI markets would stifle innovation and interoperability in a technology that threatens to upheave labor markets. Preventing LLM providers from buying up startups dependent on their software—and potentially carving up the AI applications markets—will encourage consumer choice, innovation, and the proliferation of AI applications. It will also reduce the current leading LLM model developers’ ability to monopolize distribution channels, as witnessed in the Google Search case, keeping the LLM market competitive.
Leading LLM developers have the capabilities and resources to create their own applications on top of their models. They should be encouraged to do so rather than be allowed to hoover up the most promising startups early in the life of the industry as Google, Meta and Amazon managed to do.
Surely, OpenAI has all the capabilities it needs to develop its own offering in this rapidly growing AI coding agent space. It already has the basic infrastructure with its Codex product. It has some of the best AI talent and $40 billion from a recent fundraising round. If it wants to own the coding agent market, it should have to win it rather than be able to buy it.
Finally, innovation depends not just on an individual firm’s incentives to innovate, but also on the innovative capabilities of other firms in the same or adjacent markets. A healthy ecosystem of potential competitors and partners helps generate the exchange of knowledge and opportunities for collaboration. Technological developments at one firm spur developments and efficiencies at another. This organic growth is lost when firms drop out of the market or are absorbed in larger firms. Furthermore, the Google Search case taught us that once users become familiar with one software program, they are reluctant to switch to a competing program and that these habituated user bases are critical to the competitiveness of a firm operating in the digital economy. Preserving Windsurf as an independent brand with an independent user base and the capabilities to innovate in AI will help build a more innovative AI ecosystem.
Despite all the usual arguments we can expect to hear around complementary products, synergies, and efficiencies between OpenAI and Windsurf, consumers and innovation are best served by encouraging OpenAI to compete in this new market, rather than letting it use its better access to capital to muscle its way in.
Author Disclosure: The author reports no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
