While Google And Openai Battle For Model Dominance, Anthropic Is Quietly Winning The Enterprise Ai Race

Yuichiro Chino/Moment via Getty

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways 

• Anthropic has toppled OpenAI successful stock of endeavor GenAI spending.
• Enterprise spending represents an AI boom, not a bubble.
• Agentic AI is still a niche compared to punctual engineering.

The astir celebrated azygous usage of generative artificial intelligence successful nan endeavor is coding devices for programmers. That tool's fame has helped Anthropic trounce OpenAI successful nan endeavor successful 2025, according to a caller study of institution spending.

"The instauration exemplary scenery shifted decisively this twelvemonth erstwhile Anthropic amazed manufacture watchers by unseating OpenAI arsenic nan endeavor leader," according to nan report, nan third yearly State of Generative AI successful nan enterprise, from task superior patient Menlo Ventures. 

Also: How Anthropic's endeavor power fueled its monster $183B valuation

The study estimated that Anthropic now earns 40% of endeavor LLM spend. That's up from 24% past twelvemonth and 12% successful 2023. Meanwhile, OpenAI mislaid almost half of its endeavor share, dipping to 27% from 50% successful 2023, wrote authors Tim Tully, Joff Redfern, Deedy Das, and Derek Xiao. 

Menlo Ventures

The squad based their marketplace stock connected nan dollar magnitude spent by enterprises connected accumulation API usage, arsenic weighted by nan standard of each customer. The study estimated that US enterprises spent a full of $37 billion, up much than threefold from past year's $11.5 billion.

The study is specifically based connected US spending, truthful it's not clear what nan information propose astir world trends. The squad surveyed executives successful nan C-suite, IT, and engineering astatine 495 US companies successful November, pinch nan thief of a third-party study firm, to put together an estimate for full spending. 

Anthropic dominating codification automation

The study confirms earlier findings by Menlo Ventures connected nan fame of Anthropic's Claude ample connection models, as elaborate backmost successful August. It is mostly being fueled by nan automation of code-writing.

"Anthropic's ascent has been driven by its remarkably durable power successful nan coding market, wherever it now commands an estimated 54% marketplace share, compared to 21% for OpenAI," nan study said.

Also: Anthropic thumps OpenAI arsenic nan apical LLM supplier for business - and it's not moreover close

And coding tools, represented by startups specified arsenic Replit, Cursor, Harness, Windsurf, Augment Code, and All Hands AI, are now a $4 cardinal yearly business, nan Menlo Ventures squad noted, "making it nan largest class crossed nan full exertion layer," and "Generative AI's first 'killer usage case.'"

Menlo Ventures

Anthropic's occurrence frankincense is being helped by those startups successful coding tools. That is accordant pinch anecdotal observations, specified arsenic those by package entrepreneur Jeremy Burton in my October interview. 

Also: Why AI coding devices for illustration Cursor and Replit are doomed - and what comes next

Burton pointed retired that astir coding tools, specified arsenic Cursor and Replit, dangle connected Anthropic's Claude technology. 

"Most of those startups dangle connected Anthropic's model," Burton told me. 

Menlo has a nonstop financial liking successful nan findings, arsenic it costs many startups successful nan area, including Anthropic.

An AI roar versus a bubble

Menlo Ventures

The study has a distinctly much affirmative reside than past year's study by Menlo Ventures, which elaborate nan struggles of enterprises to get AI disconnected nan ground. It appears that a large displacement successful nan intervening twelvemonth is much and much packaged applications emerging versus do-it-yourself endeavor projects.

Also: Enterprises are struggling pinch what to do pinch Gen AI, opportunity task capitalists

The authors noted that "For a while, nan prevailing contented was that enterprises would build astir AI solutions themselves." Last year, 47% of AI solutions were built internally versus 53% purchased; this year, that's reversed. 

"Today, 76% of AI usage cases are purchased alternatively than built internally. Despite continued beardown investments successful soul builds, ready-made AI solutions are reaching accumulation much quickly and demonstrating contiguous worth while endeavor tech stacks proceed to mature."

The emergence of coding tools, but besides vertical-market devices for customer acquisition and finance, prompted nan task capitalists to sound a singularly affirmative note.

Menlo Ventures

"This year's findings make clear that nan displacement is nary longer speculative," they wrote. "Enterprise AI is now a $37 cardinal marketplace -- nan fastest-scaling class successful package history. Across industries, AI has go halfway to really activity gets done. Enterprises, seeing existent returns, are doubling down."

Noting nan wide fears presently of a backing bubble successful artificial intelligence," nan authors replied that while "the concerns aren't unfounded," nan "demand broadside tells a different story," and nan "real revenue" intends that it's much "a roar versus a bubble."

The full estimated spending by enterprises this year, $37 cardinal by Menlo Ventures's reckoning, is up much than threefold from past year's number. 

Agents are still a niche

One area that truthful acold hasn't played retired is simply a class that Menlo Ventures has been rather optimistic about: agentic AI, wherever ample connection models tin person hooks into different endeavor programs and information stores to execute much precocious tasks.

So far, nan agents are not swarming nan enterprise, neither successful nan packaged applications nor successful nan AI infrastructure spending, they noted.

In nan largest and fastest-growing portion of AI applications, horizontal AI applications, almost each spending, 86%, is connected simpler co-pilot programs specified arsenic ChatGPT Enterprise, Claude for Work, and Microsoft Copilot, versus agentic AI specified arsenic Salesforce Agentforce, Writer, and Glean.

Also: The travel to afloat autonomous AI agents and nan task capitalists backing them

In nan $1.5 cardinal AI infrastructure class of spending, astir of what's being put into accumulation is not agentic; it's plain-old-fashioned punctual engineering, noted Tully and team.

"For each nan talk of 'agents,' existent accumulation architectures stay amazingly simple," they wrote. "Only 16% of endeavor and 27% of startup deployments suffice arsenic existent agents -- systems wherever an LLM plans and executes actions, observes feedback, and adapts its behaviour -- while astir are still built astir fixed-sequence aliases routing-based workflows wrapped astir a azygous exemplary call." 

The authors laid retired respective predictions astatine nan extremity of nan report, including nan imaginable that coding devices will beryllium superior to quality coders successful immoderate mundane tasks. 

"AI will transcend quality capacity successful regular applicable programming tasks," they said. "There is nary plateauing of LLM accomplishment sets, particularly successful verifiable domains specified arsenic mathematics and programming, wherever nan champion models will proceed to get amended and better."

The study whitethorn beryllium excessively bullish

It's worthy asking whether nan affirmative cognition of nan study is really supported by nan information that Tully and squad person gathered. 

A three-fold summation successful manufacture gross successful 1 year's clip is so striking, but it seems little awesome arsenic you excavation into nan numbers.

While $37 cardinal successful full yearly generative AI income successful nan US whitethorn sound for illustration a lot, it is comparatively mini compared to, for example, amounts spent connected unreality computing. This year, for conscionable nan apical 3 unreality vendors, Alphabet's Google Cloud, Amazon's AWS, and Microsoft's Azure, mixed gross is projected to scope $288 billion. 

And if you analyse nan various categories of spending compiled by Tully and team, it's clear there's very small momentum for categories extracurricular of nan astir predictable ones.

Menlo Ventures

Within nan $37 billion, $18 cardinal for infrastructure is nan worldly that has been trading steadily for 3 years now, nan usage of frontier exemplary APIs, specified arsenic Claude and GPT, and nan attendant infrastructure equipment. Within applications, $8.4 cardinal is for nan earliest devices offered, nan horizontal applications specified arsenic Microsoft's Copilot offerings, and $4.2 cardinal is for nan coding devices that are a large hit, specified arsenic Cursor. 

What that intends is that almost each of nan spending by enterprises, 83%, is simply nan astir evident usage cases, namely, renting API usage, moving co-pilots, and utilizing coding tools. It's heavy tilted toward nan method users astir inclined to prime up Gen AI, nan coders and merchandise developers. 

In almost each different class of nan remaining $6.4 billion, nan first amounts spent look minuscule. 

For example, wrong departmental AI, a full of $360 cardinal has been spent connected AI-driven quality resources applications, they noted, and $100 cardinal connected finance and operations applications utilizing AI. Those numbers are mini compared to nan yearly income of, for example, Workday, a vendor of some HR and finance applications, which is projected to make $9.5 cardinal successful yearly income this year. 

Also: ChatGPT saves nan mean worker astir an hr each day, says OpenAI - here's how

And contempt tons of talk astir trading teams successful enterprises utilizing Gen AI to switch creatives, nan full walk by trading connected Gen AI was conscionable $660 million, they estimated -- a fraction of nan gross of companies specified arsenic Adobe and Figma. 

The authors whitethorn person been a small excessively charmed by that wide three-fold complaint of maturation -- nan fastest complaint of maturation of immoderate package class successful history, arsenic they said. When an full manufacture consisting of each established package vendor connected nan planet, positive hundreds of venture-backed startups, goes aft a azygous marketplace astatine nan aforesaid time, of course, location will beryllium accelerated growth.

But erstwhile each of those companies negociate to waste only $37 cardinal successful 12 months, it's not arsenic large a triumph arsenic it seems astatine first blush.