Microsoft and Google take on Anthropic and OpenAI in AI coding models

In the booming generative AI market, Anthropic has zoomed ahead of the field, largely thanks to Claude Code, its AI coding assistant. Seeing where the money is, OpenAI shifted much of its focus from the consumer market to enterprise, where its Codex offering is going up against Claude Code.

Now, Google and Microsoft are making a concerted effort to get into the game, using their massive balance sheets and expansive cloud businesses to try and lure developers.

At its developer conference last month, Google placed heavy emphasis on agentic AI, including the unveiling of Antigravity 2.0, which the company said can “orchestrate multiple agents to execute tasks in parallel, such as having one agent code a website while another generates brand assets.” And at the beginning of a blog post announcing Gemini 3.5 Flash, Google said the new model offers “frontier performance for agents and coding.”

Microsoft’s turn on stage comes this week at its annual Build conference in San Francisco. There the company intends to announce a coding model that will be available in Copilot, emphasizing its lower price relative to alternatives, said a person familiar with the plans who asked not to be named in order to discuss internal matters.

The Information reported on the plans last week. Microsoft declined to comment.

With businesses rapidly adopting coding assistants, market research firm Mordor Intelligence predicted that the AI code tools market would expand by 26% a year, from $9.3 billion this year to roughly $30 billion by 2031. The numbers are big, but Google and Microsoft aren’t just looking for new revenue streams. They want developers using their tools, running workloads on their clouds and relying on their models so that the models themselves are able to do more training and get bigger and smarter in the process.

“It’s absolutely critical for these companies to compete in this market,” said Gil Luria, an analyst covering software and other tech stocks at D.A. Davidson.

It’s a sentiment shared by Tomasz Tunguz, founder of Theory Ventures, who described AI coding as the most attractive market for generative AI models. He estimates that AI might eventually represent 30% to 60% of research and development spending.

In addition to Anthropic and OpenAI, companies like Cursor have specifically targeted the coding opportunity. In May, Cursor signed an agreement with SpaceX that gives Elon Musk’s company the right to acquire the startup for $60 billion.

The tools have become so powerful that Wall Street started bailing on software stocks — though the beaten-down sector rebounded in May — on concern that entire companies could be wiped out because their products could be vibe coded. Anthropic on Thursday announced the closing of a financing round at a $965 billion valuation, surpassing OpenAI, and said on Monday that it’s confidentially filed for an IPO.

“AI was headed in a very consumer-centric direction, but Anthropic also became laser-focused on coding because they had this understanding that coding is where the AI frontier was going to be,” Luria said. “When everyone else was getting distracted by images and videos, Anthropic knew coding was going to drive the performance of models and help with everyone else’s tasks, and they’re all now pivoting to coding.”

Also last week, Anthropic unveiled an upgrade to Claude Opus, its most advanced model for complex tasks like coding. With Opus 4.8, Anthropic announced a new 1 million token default context, raising the amount of working memory it offers for complex coding tasks in a single conversation.

Google, meanwhile, reset Gemini’s token quota for its Antigravity product following its developer conference last week, raising model rate limits after developers complained that they used up initial quotas too quickly.

And Microsoft is starting to charge for its Copilot AI coding assistant based on usage to align more closely with rising costs.

“Anytime you ask any of these tools, ‘Build this thing for me,’ they’re burning tokens,” said Ken Parmelee, an analyst at technology industry researcher Forrester. “This is the new gateway drug. It’s a way to hook people into their other products.” A token is equal to about three-quarters of one word.

At data analytics software company Snowflake, programmers mainly depend on the company’s own CoCo development tool, as well as Claude Code, CEO Sridhar Ramaswamy said. He said it’s not uncommon for a “phenomenally productive engineer” to spend $50,000 a year on it.

In addition to Antigravity, Google has Gemini Code Assist that can work inside a variety of code editors. Last year, Google signed a $2.4 billion licensing deal for Windsurf’s technology and hired the AI coding startup’s CEO, Varun Mohan, along with key researchers.

Google CEO Sundar Pichai acknowledged his company’s challenge in coding in a recent interview with the Hard Fork podcast.

“When it comes to agentic coding with tool use, and instruction following, long-horizon tasks, I think we are a bit behind at this moment,” Pichai said

At Google I/O, the company marketed itself as a more affordable option for coders, announcing an AI developer subscription tier at $100 per month.

“They can afford to give away the tools cheaper, because they know if you’re in their ecosystem, you’re going to pay to use their memory, their integrations,” Parmelee said. He added that the tense competition gives Google an opportunity to elbow its way in, calling it “a very tumultuous market right now.”

“The more users they have on the platform, and the more use cases, the better quality the models will be, which means the better the outputs will be,” Parmelee said.

While DeepMind serves as Google’s AI lab, Luria is quick to point out that Google has a massive business offering cloud infrastructure and services.

“They are in the business of providing enterprises with tools,” Luria said.

Even with Anthropic jumping out to a big lead and OpenAI racing to catch up, users of coding tools are regularly experimenting with multiple options, and there’s very little vendor lock-in. Database software maker MongoDB, for example, has rolled out three generative AI tools, including Anthropic’s Claude Code, to its engineers, CEO CJ Desai said.

“For different use cases we use different things,” he said. “So if we find that one solves all of them, of course I would like to consolidate.”

MongoDB is buying AI products one year at a time rather than opting for longer deals, said Desai.

“If Gemini came up with something better, or Codex is better, then I want to be able to use that and not do a long-term commitment, right?” he said.

Desai said that AI coding tools currently represent “a very small fraction of our people cost.” But if the cost goes up substantially, he said, the company would impose limits.

The pressure is on for Microsoft to stay in the game. The software giant has direct access to millions of developers through coding repository GitHub. And through GitHub Copilot, developers can tap models from Anthropic, Google and OpenAI.

Whether Microsoft will simply offer all of the top coding models or will also be a provider of a key model is an open question.

On Microsoft’s January earnings call, finance chief Amy Hood talked about the company’s need to balance its AI infrastructure requirements across Azure, products such as GitHub Copilot and research that can bring about product innovation.

“Build presents opportunities for perception there to change if they can show things that are more proprietary to Microsoft,” said Michael Turrin, an analyst at Wells Fargo who recommends buying the stock.

Developers will generally gravitate toward the newest and most capable AI models, Turrin said.

“Microsoft is going to have to give some specific use cases beyond just cost for that audience,” he said.

When GitHub Copilot launched in 2021, relying exclusively on OpenAI, it was a pioneer in the nascent AI coding space. But it’s lost momentum with the speedy emergence of Cursor and Claude Code.

Cursor, the AI-powered code editor, has 300 employees and is one of the fastest-growing cloud software companies in history, going from $4 million to $2 billion in annualized revenue in just 18 months. With the potential SpaceX deal, “People are like, I’m not sure what to do with this, but it’s still a great tool,” Parmelee said.

Rob Sanfilippo, an analyst at advisory firm Directions on Microsoft, said Microsoft “might be able to really differentiate by building their own models with what they’ve learned about the capabilities of the models they’ve been competing with, and in some cases losing a lot of their early share to.”

Sanfilippo said this could be part of what CEO Satya Nadella “has planned for correcting the course on AI at Microsoft.”

WATCH: AI Tokens or humans? The new debate reshaping corporate budgets