Artificial intelligence coding agents provide enormous value to consumers for very low fees. But the market is quickly shrinking with Anthropic in the lead. Only competition, and requiring Big Tech to build agents rather than buy them, will continue to let AI’s value flow to consumers. As such, the courts should ban SpaceX’s recently proposed acquisition of Cursor, writes Ketan Ahuja.
Anyone who’s used Anthropic’s Claude Code knows how powerful artificial intelligence, and specifically AI coding agents, the software programs that can write code and execute tasks using large language models, have become. The market for coding agents, is growing so fast that it’s hard to size accurately, but predictions suggest it may grow from around $7.63 billion in 2025 to around $183 billion in 2033. [worth $6.4 billion in 2025, is expected to burgeon to nearly $100 billion by 2034.] It promises to reduce the time it takes software engineers to complete a coding or debugging task from hours, sometimes even days, to minutes. The sheer advances in productivity will spur efficiencies, savings, innovation, and business formation.
This means the companies that make good AI agents like Anthropic have enormous pricing power. If an AI agent can code like a software engineer, then AI companies can charge as much as a software engineer, which is hundreds of thousands of dollars. What is keeping Anthropic, the market leader, from raising its prices is competition. The reported acquisition of Cursor, another sophisticated coding agent, by Elon Musk’s SpaceX, underlines that competition in this market is decreasing. If the market is allowed to consolidate, prices will rise in the long term.
The state of the market
Cursor initiated the coding agent market in 2023, followed by Devin (March 2024), Windsurf (November 2024), Anthropic’s Claude Code (February 2025), OpenAI’s Codex (April 2025), xAI’s Macrohard (August 2025), and Google’s Antigravity (November 2025). OpenCode is an open-source coding agent that is of lower quality but is a viable offering for sophisticated users. There may be other marginal players that don’t claim significant market share or have much visibility. Some AI companies also compete in parts of this coding agent market, such as website-builder Lovable, and AI agent platform OpenClaw, but are not full-service coding agents.
Good data on adoption is hard to come by as these are private companies, but Claude Code seems to be the clear winner, as it has powered Anthropic to an estimated $30 billion in annual recurring revenue for 2026, up from $9 billion in 2025. Codex and Cursor seem to be in second and third place, with Cursor at $2 billion in annual revenue. Each of these products has among the fastest growing revenue of all time. The others seem to have lower adoption.
Quality really matters for a coding agent: if the agent can complete long-running tasks without supervision and get things mostly right, it can make software engineers much more productive. If it makes a lot of mistakes, it becomes a time sink. Anthropic’s Claude Code hit the right quality standard at the start of the year, becoming the first agent to operate autonomously and get things mostly right.
The race for dominance
This market is already consolidating, with everyone chasing Anthropic. Last year, OpenAI tried to buy Cursor and Windsurf, before Google hired Windsurf’s founders and licensed its intellectual property for $2.4 billion in what is called a reverse acquihire. The rest of Windsurf merged with Devin. OpenAI hired OpenClaw’s founder for an undisclosed sum. SpaceX has now struck a deal to buy Cursor for $60 billion and combine it with xAI.
These acquisitions are attempts for competitors to buy market leadership using their firehose of money rather than careful execution. Elon Musk’s attempt to build a coding agent has gone slowly alongside the broader chaos at xAI, where all of its 11 cofounders recently left. OpenAI’s attempt to compete across all AI product verticals meant it lost focus. OpenAI and xAI are still in the race though: they have the capital and talent to improve their existing coding agents without having to buy up competitors. The market is still very young, and Anthropic’s hold is far from assured.
Big Tech’s consolidation throughout the 2000s and 2010s provides the clearest warning about allowing the coding agent market to consolidate through mergers and acquisitions. The Federal Trade Commission just unsuccessfully attempted to unwind Facebook’s purchases of Instagram and WhatsApp in 2012 and 2014, respectively, which allowed it to cut off competition to its social media empire. The Department of Justice did succeed in alleging Google had monopolized the internet search and ad tech markets, in part aided by its acquisitions of market players like DoubleClick and ancillary platforms like Android.
When firms compete for market leadership, as Google did when it launched its search engine, consumers benefit from cheaper, higher quality, and more innovative goods. The market only benefits when firms are forced to build and compete.
Which is why the courts should block under antitrust law SpaceX’s acquisition of Cursor and any other similar future acquisitions proposed in the future in this nascent coding agent market—and perhaps the broader AI market as well. When SpaceX or OpenAI attempt to buy Cursor or Google hires away the founder of a competitor, leaving behind a firm bereft of the employees who are the firm’s intellectual property and most important competitive edge, it deprives the market of the competing products that SpaceX, OpenAI, and Google would develop. Instead, they choose the simpler and more strategic route. Not only do these Big Tech firms buy their way into the market, but they remove a competitor with the same move.
We’re already in the danger zone in terms of the number of companies that offer good coding agents. When there are only a few firms in a market—and if SpaceX is allowed to buy Cursor, we’ll be down to five quality products—it becomes easier for firms to mirror each other’s prices, and markets settle into a stable, high-price oligopoly. This is starting to happen: all AI companies seem to charge similar, coordinated prices, with subscription tiers at $20, $100, and $200, though there is still price competition as the same headline price gets you more tokens with some providers. If there are only three or four good AI coding agents, the companies that make them will be able to coordinate tacitly on higher prices.
These high prices won’t attract new firms to enter the market if the current AI companies are allowed to consolidate. Companies need enormous amounts of capital and distribution channels to build effective coding agents, with each of the companies listed above raising hundreds of millions or even billions of dollars. Newcomers without capital and distribution will struggle to get started: they would be competing with good incumbents, and no VC would fund them. The incumbents will also have a significant head start on accumulating the data and signals that allow for quality AI products. The likely set of capable AI coding agents going forward includes the list above, and any Big Tech firms with the resources and distribution channels to muscle their way in.
Antitrust laws once intervened early to stop a market from consolidating into an oligopoly through mergers, seeing mergers like this for what they really are. Sadly, antitrust laws are less able to block these mergers today. President Donald Trump sold antitrust policy to his wealthy Big Tech donors, and an embedded free-market ideology among the courts has led them to ignore market realities and see these mergers as mostly benign. Bold state attorneys general and citizen groups will have to step in to enforce antitrust and convince the courts to look through the ideological halitosis that fast-talking lawyers will throw at them.
The United States is living through the consequences of letting Big Tech platforms roll up their industry in the 2000s and 2010s. It’s critical that we avoid repeating the same mistakes with AI. If Elon Musk wants a piece of this market, he should use his resources to out-innovate his competitors rather than buy them up.
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.
Subscribe here for ProMarket’s weekly newsletter, Special Interest, to stay up to date on ProMarket’s coverage of the political economy and other content from the Stigler Center.
