Who’s buying AI coding startups?
In our AI code assistant market deck, you will find everything you need to understand the market
SUMMARY
AI coding startups are being bought most often by other AI coding platforms, while the biggest checks are coming from giant technology companies such as SpaceX, Google and Nvidia.
The buyer landscape is split. Cursor, Cognition and Anaconda have been repeat acquirers, using M&A to fill specific product gaps, while Big Tech has moved less often but with much larger and sometimes unconventional transactions.
A surprisingly important shift is that buying an AI coding startup no longer always means buying the company. Google’s Windsurf transaction and Nvidia’s Poolside agreement show that technology licenses, investment and mass hiring can transfer much of the strategic value without a conventional acquisition.
The market is also moving beyond the coding interface itself. Buyers are now targeting code review, secure execution environments, build infrastructure, production monitoring and operational tooling, which suggests the valuable battleground is becoming the entire software-development loop.
That helps explain why OpenAI and Anthropic have bought infrastructure such as Ona, Promptfoo and Bun rather than simply adding another editor. Once the coding agent already has distribution, execution reliability, testing, security and enterprise governance become harder bottlenecks.
Code review is emerging as one of the clearest pressure points. If AI causes engineering teams to produce several times more code, the economic value shifts toward tools that can inspect, test and manage those changes without adding the same amount of human review work.
Microsoft is the notable exception to the acquisition rush. GitHub, VS Code and Copilot already give it distribution across developers, so it can let competing agents operate through its platform rather than paying tens of billions to recreate an audience it already owns.
The universe of possible buyers is widening as well. Figma and Cloudflare show that design platforms, infrastructure vendors and companies adjacent to software creation can become credible acquirers when AI coding starts to overlap with their core product.
The strongest independent AI coding companies are becoming difficult to acquire at normal software prices. Cognition, Lovable and Replit now carry multibillion-dollar valuations and can raise huge private rounds, which pushes many buyers toward smaller capability deals, teams and infrastructure assets instead.
The next deals will probably be uneven: frequent tuck-ins for very specific capabilities, plus occasional multibillion-dollar technology or talent transactions when a scarce team becomes strategically important. The real M&A race is increasingly about who controls the bottlenecks between a developer’s request and code that safely reaches production.
Who is actually buying AI coding startups right now?
Right now, AI coding companies themselves are the most frequent buyers, while SpaceX, Google and Nvidia are behind the biggest disclosed transactions.
We found a market with two very different kinds of buyers. Cursor made at least four clear acquisitions before becoming part of SpaceX, buying Supermaven, Graphite, Continue and Firetiger. Cognition bought Windsurf and has since kept shopping around the software-agent stack. Anaconda bought Kilo Code. These companies are using M&A to widen what their coding products can do.
The much larger technology companies are behaving differently. SpaceX bought Cursor for $60 billion. Google paid roughly $2.4 billion for Windsurf technology rights while hiring its CEO, co-founder and researchers. Nvidia has just agreed to pay $6 billion to license Poolside technology, invest another $1 billion in the company and hire more than 100 employees. OpenAI bought Ona, while Anthropic bought Bun.
Then there is a third group starting to appear: companies that already sit next to software development and want to move closer to actual code creation. Figma acquired the Bud team, formerly Orchids, while Cloudflare bought VoidZero, the company behind Vite.
| Buyer | AI coding or developer target | How the deal worked |
|---|---|---|
| Cursor | Supermaven, Graphite, Continue, Firetiger | Acquisitions |
| Cognition | Windsurf | Acquisition |
| Windsurf technology and senior team | ~$2.4B license and hiring deal | |
| OpenAI | Ona | Acquisition |
| Anthropic | Bun | Acquisition |
| Anaconda | Kilo Code | Acquisition |
| SpaceX | Cursor | $60B acquisition |
| Nvidia | Poolside technology and staff | $6B license, $1B investment, hiring |
| Figma | Bud / Orchids team | Team acquisition |
| Cloudflare | VoidZero | Acquisition |
Does “buying an AI coding startup” still mean buying the company?
Increasingly, no: some of the biggest AI coding deals today transfer the engineers and technology while leaving the startup itself alive.
Google's Windsurf transaction is the cleanest example. OpenAI had been discussing a roughly $3 billion acquisition of Windsurf, but that offer expired. Google then paid about $2.4 billion for non-exclusive technology rights and hired CEO Varun Mohan, co-founder Douglas Chen and several senior researchers. Cognition subsequently bought the remaining Windsurf product, IP, brand and business.
Nvidia has now gone even bigger with Poolside. The Wall Street Journal reported that Nvidia is paying $6 billion for a non-exclusive technology license, investing another $1 billion at a $12 billion pre-money valuation and offering jobs to 109 Poolside employees. Poolside remains an independent company.
OpenAI also showed how far the definition can stretch when it hired at least seven employees from open-source coding startup Cline. OpenAI explicitly said there was no acquisition, and Cline remains operational. More recently, Google has reportedly been discussing another license-and-hire deal worth more than $1.5 billion with Mechanize, a small startup working on software-engineering automation and training environments.
We should therefore separate three things throughout this market: buying the company, buying the technology, and buying the people. All three can remove scarce capabilities from a competitor, even though only the first one appears in a conventional M&A database.
This market map, featured in our AI code assistant market deck, highlights top companies and startups in the AI code assistant market
Is Big Tech really driving the AI coding acquisition wave?
Only partly: Big Tech currently dominates the dollars, but AI-native companies have been much more repetitive buyers.
Before its sale, Cursor completed four clear acquisitions across roughly two years. Cognition has bought three companies in roughly a year, starting with Windsurf and later adding TierZero and The Interaction Company, which makes Poke. Anaconda has also been assembling an AI development platform through multiple deals, including Kilo Code, Outerbounds and Enkrypt AI.
Most of the giants have been less frequent. Google has one completed multibillion-dollar coding-related talent and licensing transaction and another reportedly under discussion. Nvidia's newest Poolside agreement is enormous, but it is still one transaction. Anthropic has made one especially relevant coding acquisition in Bun. OpenAI has been active around agents and coding infrastructure, although it has usually bought specific capabilities rather than another major coding assistant.
Microsoft is the striking omission. GitHub Copilot already had 4.7 million paid subscribers, up 75% year over year, according to Microsoft's fiscal second-quarter earnings call. By the following quarter, nearly 140,000 organizations were using Copilot and enterprise subscriptions had nearly tripled. Microsoft has enough distribution to keep building instead of buying every new coding interface that becomes popular.
For now, this looks more like a young industry consolidating itself than a Big Tech land grab, with occasional giant transactions when a large technology company decides an asset has become strategically unavoidable.
If you want more recent data on this point, please see our latest AI code assistant market report.
Why are OpenAI and Anthropic buying infrastructure instead of another coding app?
OpenAI and Anthropic currently seem more interested in fixing the weak points around their coding agents than in buying another editor.
Anthropic bought Bun after Claude Code had already become a major business. Bun combines a JavaScript runtime, package manager, bundler and test runner, and Anthropic had already been using it internally. When the acquisition was announced, Claude Code had crossed $1 billion in run-rate revenue just six months after its public release. Anthropic later disclosed that Claude Code had passed $2.5 billion in run-rate revenue and that enterprise customers accounted for more than half of it.
Once the coding agent itself is growing that quickly, buying another front-end product becomes less urgent. Reliability, execution speed and the surrounding development environment start to matter more.
OpenAI's Ona deal follows the same logic. Ona, previously Gitpod, gives Codex secure cloud environments where agents can keep working for hours or days without being tied to the developer's laptop. OpenAI said Codex had passed 5 million weekly users when it announced the acquisition, up roughly 400% from earlier in the year.
OpenAI also bought Promptfoo, whose security and evaluation tools were used by more than 25% of Fortune 500 companies according to the startup. Promptfoo is being integrated into OpenAI Frontier rather than exclusively into Codex, but the direction is similar. As agents do more real work, buyers need execution, testing, security and governance around the model.
As this chart shows, and as featured in our AI code assistant market deck, search interest in AI code assistants has increased significantly
What parts of the AI coding stack are getting bought now?
AI coding M&A is spreading beyond code generation into review, deployment, infrastructure and the messy work that begins after an agent writes the code.
Cursor's move into production operations is especially telling. Firetiger built agents that watch software after deployment, catch regressions and investigate incidents. Cursor said the acquisition should eventually let a coding agent ship a change, observe how the software behaves in production and react when something breaks.
Cognition has gone after a similar problem through TierZero, which focused on automating operational work around running software. Anaconda approached the stack from another direction by acquiring Kilo Code, an open-source, model-agnostic coding platform used by more than 3 million developers, after previously buying workflow-orchestration company Outerbounds.
The infrastructure companies are moving too. Cloudflare's VoidZero acquisition brought the team behind Vite, Vitest, Rolldown and Oxc into Cloudflare. Vite alone was seeing more than 130 million weekly downloads when the deal was announced. Cloudflare explicitly connected the acquisition to a world where autonomous coding agents create and ship much more software.
| Part of software development | Recent acquisition | What the buyer gains |
|---|---|---|
| Code generation | Cursor + Supermaven | Coding models and completion technology |
| Code review | Cursor + Graphite | Review and pull-request workflow |
| Open coding agents | Cursor + Continue | Open-source agent technology and community |
| Production operations | Cursor + Firetiger | Deployment monitoring and incident feedback |
| Enterprise operations | Cognition + TierZero | Automation after software ships |
| Agentic coding interface | Anaconda + Kilo Code | Direct access to 3M+ developers |
| Build infrastructure | Cloudflare + VoidZero | Vite and the surrounding JavaScript toolchain |
Why is AI code review suddenly so valuable?
AI code review is becoming a hot acquisition category because coding agents are creating code faster than engineering teams can comfortably inspect it.
Cursor bought Graphite after Graphite had last been valued at $290 million. TechCrunch reported that Cursor paid "way over" that previous valuation. Graphite gave Cursor a mature review workflow and a product already used by hundreds of thousands of engineers.
The fresher evidence comes from CodeRabbit. The AI code-review company has just raised $143 million at a $1.5 billion valuation after growing revenue more than fivefold year over year. CEO Harjot Gill told Axios that customers are now seeing roughly three to five times more code generated as AI spreads through engineering teams.
If a company suddenly produces four times as much code, forcing humans to review everything manually becomes painful even if the generated code is reasonably good. Review, testing and change management get more valuable as generation gets cheaper.
CodeRabbit is already pushing beyond reviewing individual pull requests into what it calls "Agentic Change Management." Gill also told Axios that the company has started looking at acquisitions of its own. The code-review layer may therefore produce both acquisition targets and another new buyer.
If you want more recent data on this point, please see our latest AI code assistant market report.
This chart, featured in our AI code assistant market deck, illustrates yearly VC funding for AI code assistant startups
Why did SpaceX pay $60 billion for Cursor?
SpaceX paid $60 billion for Cursor because AI coding has become a huge application business and Cursor gives SpaceX a direct way to turn massive computing capacity into something millions of developers can use.
The deal would make little sense if Cursor were still just a clever VS Code fork. By the time SpaceX moved, Cursor was training its own models, selling heavily into enterprises and turning the editor into a broader agent platform. Cursor said earlier this year that more than 70% of the Fortune 500 were using the product.
The financial context is useful too. Before the acquisition, Cursor had been preparing to raise about $2 billion at a valuation around $50 billion, according to TechCrunch. SpaceX's $60 billion offer was therefore only around 20% above the valuation Cursor was already trying to establish privately. The headline number is enormous, but the acquisition price was much closer to Cursor's latest financing trajectory than it first appears.
The deal has now closed. Cursor says being inside SpaceX gives it access to what it describes as the world's largest fleet of GPUs, which it can use to train stronger coding models and reduce inference costs. SpaceX gets the opposite side of that equation: a mature application and developer distribution channel for all that compute.
This is probably the clearest example yet of AI coding becoming strategically important enough to justify a mega-cap acquisition rather than an ordinary software tuck-in.
Why did Cognition buy Windsurf after Google took its founders?
Cognition bought Windsurf because the remaining company still had a surprisingly large business even after Google hired away its most important research leaders.
According to Cognition's acquisition announcement, Windsurf still had $82 million in annual recurring revenue, more than 350 enterprise customers, hundreds of thousands of daily active users and enterprise ARR that had doubled quarter over quarter. Cognition acquired the product, IP, trademark, brand and remaining team.
That combination is unusual. Google had already removed CEO Varun Mohan, co-founder Douglas Chen and several senior researchers through its separate licensing and hiring agreement. Yet the commercial shell left behind was still valuable enough for another AI coding company to buy.
Cognition effectively bought something that would have taken years to recreate: an IDE people already knew, a sizeable enterprise customer base, sales relationships and another route into developers' daily work. Devin could remain the autonomous engineering agent while Windsurf gave Cognition a more traditional interactive coding surface.
The deal shows why "talent versus company" has become such an important distinction in AI M&A. A startup can lose much of its frontier research team and still contain tens of millions of dollars of recurring revenue, thousands of customer relationships and useful distribution.
If you want more recent data on this point, please see our latest AI code assistant market report.
This chart, featured in our AI code assistant market deck, breaks down Anyshpere’s playbook in AI code assistants
Why isn't Microsoft buying AI coding startups?
Microsoft currently has less reason than almost anyone else to buy its way into AI coding because it already owns GitHub, VS Code distribution and one of the biggest paid coding assistants in the world.
GitHub Copilot had more than 4.7 million paid subscribers at Microsoft's fiscal second-quarter update. A quarter later, Microsoft said nearly 140,000 organizations were using Copilot and enterprise subscriptions had almost tripled year over year. GitHub itself had already passed 180 million developers.
Microsoft is also turning GitHub into a place where competing agents can operate instead of trying to own every model. Agent HQ was designed to let developers manage agents from OpenAI, Anthropic, Google, Cognition, xAI and others through GitHub's existing pull requests, issues and actions.
Microsoft reportedly looked at buying Cursor before the SpaceX transaction but never made an offer. That decision makes more sense when we look at the assets Microsoft already has. Buying a $50 billion-plus coding company would duplicate a large amount of its existing developer distribution.
Microsoft can still make an acquisition if a competitor creates something it cannot copy quickly enough. For now, GitHub gives Microsoft the luxury of watching the consolidation without having to participate in every deal.
Are Figma and Cloudflare becoming AI coding buyers too?
Yes, companies that used to sit next to coding are now buying their way directly into software creation.
Figma's acquisition of the team behind Bud, previously called Orchids, is a clean example. Bud started as a vibe-coding product that could create web and mobile applications from prompts and later expanded into a broader agent capable of browsing the web, accessing services and writing code. After the deal, the standalone Bud and Orchids products were shut down.
Figma already had Figma Make and integrations with Codex and Claude Code. Bringing in an AI coding team pushes the company further from "design software" toward a place where a user can go from an idea to a working application without leaving the broader Figma environment.
Cloudflare is coming from the other side of the workflow. Its acquisition of VoidZero gives it the team behind Vite and several other core JavaScript tools. Cloudflare said the goal is to make the path from locally generated code to production infrastructure much easier for both developers and autonomous agents.
These buyers widen the acquisition universe. An AI coding startup no longer needs Microsoft, Google or another coding startup as its natural buyer. Design platforms, cloud infrastructure companies, observability companies, cybersecurity vendors and data platforms can all have a reason to own part of the software-creation workflow.
This chart, featured in our AI code assistant market deck, illustrates yearly funding for AI code assistant startups
Are the best AI coding startups already too expensive to buy?
For most potential acquirers, yes: the strongest independent AI coding companies are now worth billions to tens of billions of dollars.
Cognition's latest financing valued the company at about $26 billion post-money after it raised more than $1 billion. Cognition said its annualized revenue run rate had reached $492 million, while enterprise usage of Devin had grown more than tenfold since the start of the year.
Lovable's newest round valued the company at $13.3 billion after raising another $400 million. Replit raised $400 million at a $9 billion valuation and says its products are now used somewhere inside 85% of the Fortune 500. Factory raised $150 million at a $1.5 billion valuation after saying revenue had doubled month over month for six consecutive months. CodeRabbit has just reached the same $1.5 billion valuation.
Those prices change who can buy whom. A $100 million coding startup can plausibly be acquired by dozens of software companies. A $10 billion company requires a much smaller set of buyers, and a $25 billion company is already in mega-deal territory.
The funding market is also giving founders a real alternative to selling. Cognition can raise more than $1 billion privately. Lovable can raise $400 million. Replit can do the same. Strong companies no longer need an acquirer simply to finance the next stage of growth.
That pushes more M&A toward smaller products, specific teams and infrastructure layers where a buyer can still obtain an important capability without spending $10 billion.
If you want more recent data on this point, please see our latest AI code assistant market report.
Which AI coding startup could actually get bought next?
Mechanize is the clearest live candidate today because Google is reportedly already discussing a deal worth more than $1.5 billion for its technology and part of its team.
Mechanize is particularly interesting because it works on environments where AI systems learn to perform software-engineering tasks rather than simply providing another coding interface. Business Insider reported that the startup had previously raised only about $9 million and was valued around $500 million, making the reported price under discussion unusually high relative to the capital invested.
Cline is another company worth watching, although the evidence points to continued independence for now. OpenAI hired at least seven Cline employees for Codex earlier this year, but both sides said there had been no acquisition. Cline has since continued developing its open-source agent and currently says it has more than 8 million developers across its platforms.
The more interesting pool may eventually be companies one layer away from the coding agent. Review, testing, CI, developer environments, security and production monitoring are increasingly valuable as autonomous agents write more of the code. CodeRabbit's recent $1.5 billion financing shows how quickly those businesses can become expensive too.
We would expect the next transactions to split into two camps: relatively small capability acquisitions that fill a precise hole in a larger platform, and occasional billion-dollar talent or technology deals when Google, Nvidia or another giant decides a research team is too important to leave independent.
This chart, featured in our AI code assistant market deck, compares the main business model options for AI developer tools platforms
So who's buying AI coding startups?
Right now, the strongest answer is that AI coding platforms are buying most often, while a handful of giant technology companies are paying far more when they decide to move.
The acquisition pattern has become much clearer lately. Successful coding platforms want to own more of the software-development loop, so they are adding review, execution, infrastructure, operations and new agent interfaces. Frontier AI companies are buying the pieces that make their own coding agents work reliably inside companies. Cloud, design and developer-platform companies are now moving in from the edges.
At the very top of the market, ordinary acquisitions are becoming harder. The leading independent AI coding companies can raise hundreds of millions or billions of dollars and carry valuations that rule out almost every conventional software buyer. That is why we are seeing so many unusual licensing, talent and partial-technology deals alongside normal acquisitions.
The AI coding M&A race is broader than a fight over who owns the best code editor. Buyers are trying to control the path from a developer's request to generated code, review, execution, deployment and production feedback. These days, the most valuable acquisition target may be whichever company owns the next bottleneck in that chain.
If you want more recent data on this point, please see our latest AI code assistant market report.
OUR METHODOLOGY
The answer to “Who’s buying AI coding startups?” looks obvious until we separate the deals that attract headlines from the buyers that are actually acquiring repeatedly. We therefore looked at acquisition frequency, deal value, transaction structure, buyer type and the parts of the software-development stack being targeted.
Completed transactions and first-hand company disclosures received the most weight. Reported negotiations, financing data, operating metrics and credible reporting were used when they helped test whether a pattern extended beyond one transaction. We kept confirmed acquisitions separate from reported discussions and from license-and-hire structures.
We also treated acquisition frequency and acquisition value as different questions. A company completing several smaller deals can be a more active consolidator than a giant technology company making one multibillion-dollar move, so neither measure was allowed to stand in for the other.
The analysis also includes technology licensing and talent transactions where they materially transfer capabilities from an AI coding startup, because deals such as Google–Windsurf and Nvidia–Poolside show that strategic control can move without the startup itself being fully acquired.
Key sources used for this analysis include Cursor on Supermaven, Cursor on Graphite, Cursor on Firetiger, Cursor on joining SpaceX, Cognition on Windsurf, OpenAI on Ona, Anthropic on Bun, OpenAI on Promptfoo, Anaconda on Kilo Code, Cloudflare on VoidZero, TechCrunch on Figma and Bud / Orchids, Axios on CodeRabbit, TechCrunch on Cursor’s financing trajectory, Business Insider on Google and Mechanize, The Wall Street Journal on Nvidia and Poolside, and Microsoft’s latest GitHub Copilot disclosures.
This chart, featured in our AI code assistant market deck, illustrates how market revenue is distributed across customer segments in the AI code assistant market
Related blog posts
- Vibe coding: which startup is ahead?
- Who is winning AI coding now?
- The startups that have raised the most funding in the AI code assistant market
- Are AI coding agents too expensive for now?
- Are AI coding agents just a temporary trend?
- The most recent funding news in the AI code assistant market
