Anthropic Overtakes OpenAI: What the $8.4B Enterprise AI Market Reveals

The balance of power in enterprise AI has shifted quietly but decisively. According to a mid-2025 market study by venture firm Menlo Ventures, Anthropic — the maker of the Claude assistant — has overtaken OpenAI, the company that lit the fuse with ChatGPT, for the top spot in business usage.

The numbers are striking. Annual enterprise spending on large language models (LLMs) more than doubled in just six months, from roughly $3.5 billion in late 2024 to $8.4 billion by mid-2025. In that fast-growing market, Anthropic led with 32% of enterprise usage, while OpenAI — which once commanded about half the market — slipped to 25%, with Google close behind at 20% (all figures Menlo Ventures, mid-2025).

This reversal is not a passing fad. It reflects a change in what companies now want from AI, and that change in priorities is what reshuffled the rankings. This piece unpacks the conditions behind the "AI that gets chosen," and how both buyers and investors can put the shift to use.

The Numbers Behind the Reversal

Start with the facts. Menlo Ventures' study draws on interviews with 150 technical leaders, from AI startups to large enterprises. What matters is that it tracks workloads actually running in production — not a wish list of what people would like to try. It is not a popularity poll; it shows where budgets are actually flowing.

*Production enterprise LLM usage share. Based on Menlo Ventures' "2025 Mid-Year LLM Market Update" (as of mid-2025).*

The spend figure is just as notable. Doubling from $3.5 billion to $8.4 billion in six months (late 2024 to mid-2025) means companies promoted AI from "something to try" to "a tool that gets a real budget line." The report notes that compute usage has shifted from training (teaching the model) to inference (actually using it in operations), pushing the market into a phase where practical performance dictates vendor choice.

In short, it is not simply that wallets opened. The way the money is spent swung hard toward production use — and that is the real engine of this market's growth.

Why the Rankings Flipped: From Smart to Practical

So why did the later entrant, Anthropic, reach the top? The key is that the yardstick companies use moved from "which model is smartest?" to "which can we keep using safely in real work?"

Software development is the telling example. Claude earned strong marks for writing and fixing code and became a daily tool for engineering teams — a shift widely credited with lifting Anthropic's enterprise share. It was not the glamour of intelligence but quiet, dependable usefulness in everyday work that decided the choice.

On top of that, what enterprises dislike most is downtime and unpredictability. Response stability, handling of confidential data, ease of permissions and audits — these unglamorous operational requirements now weigh more heavily than scores on an intelligence test. OpenAI remains a major force, but its runaway lead from the days when it held half the market is over; several players now compete on merit.

Multi-Model and Governance Became the Default

Into 2026, the trend grew even clearer. A telling sign: rival Google, at Google Cloud Next '26 in April, loaded 200-plus models — including Anthropic's Claude — onto its new enterprise platform. Rather than locking customers into one company's model, "multi-model" usage, choosing the best model per task, became the baseline assumption on the buyer side.

Behind this is a rising demand for governance. Data-platform giant Snowflake partnering with Anthropic to make Claude usable in a controlled environment fits the same pattern. Being able to manage and audit who accessed which data and what was produced — that assurance now matters in production as much as performance does.

This change also confronts AI vendors with a harsh reality. Shipping a one-off "best model" is not enough: lacking the connectivity, control, and cost visibility enterprises want, share can move quickly. The flip side is that even a latecomer can leapfrog if it is strong on this ground — and Anthropic's rise proves it.

How Buyers and Investors Should Read This

For companies weighing adoption, the lesson is clear. Do not pick a vendor solely on "who is smartest right now." In a market where the leader changes within six months, today's best may not be next year's. Securing the ability to switch matters more for long-term leverage and cost control. At selection time, check whether the platform can swap between multiple models and whether it rides common interoperability standards.

It is just as important to spell out your evaluation criteria internally. Is what truly helps your business response accuracy, data control, or cost predictability? Drop in a "well-reviewed model" without deciding that, and requirements and reality will drift apart in operation. The discipline required is to choose a vendor backward from your own business needs, not from the rumor of cleverness.

From an investor's vantage point, the very fluidity of this market is instructive. If share can swing sharply in six months, the winner is not yet fixed. Rather than reacting to headlines about valuations or model benchmarks, it is a more reliable long-run signal to watch how well each company is building the machinery to keep getting chosen in production — connectivity, control, and developer support.

What to Watch in the Second Half of 2026

One caution is in order. The central figures here are from the mid-2025 Menlo report. Because the AI market's map can move within half a year, these shares should not be treated as fixed. What to read is not the snapshot of "what percent is Anthropic" but the direction of the structural shift: from a contest of intelligence to a contest of operations.

The focus for late 2026 will likely be the cost and security of running agents — autonomous AI — in earnest. The more autonomous AI you deploy, the harder spending is to forecast and the more there is to manage. Here too, the deciding factor looks less like cleverness and more like the operational ability to run things cheaply, safely, and without interruption.

The map will keep moving. That is precisely why the most prudent stance is not to bet everything on one vendor, but to make flexibility — support for multiple models and common standards — a core principle of your strategy.

Key takeaways

In Menlo Ventures' study (as of mid-2025), Anthropic led the enterprise LLM market with a 32% usage share, overtaking OpenAI (25%), with Google at 20%. The market itself doubled from $3.5B to $8.4B in six months (late 2024 to mid-2025). The reversal stems from the evaluation axis shifting from "intelligence" to "operations, governance, and connectivity in real work." By 2026, multi-model and governance had become the default, and even Google loaded Claude onto its own platform. Companies should avoid locking into a single vendor, preserve the option to switch, and choose backward from their business requirements. Because share is fluid, reading the direction of the structural shift matters more than any snapshot number.