Anthropic overtakes OpenAI in the enterprise: don't bet on just one

This last week of August 2026 has produced two signals worth reading together if your company depends on AI to get work done: Anthropic has overtaken OpenAI in enterprise LLM spend, according to Menlo Ventures’ mid-2026 report, and Barret Zoph, the person OpenAI put in charge of its enterprise sales, has left the role barely five months after taking it to return to Google DeepMind. At Evicron, an AI and custom software studio based in Barcelona, we read it as the same message twice over: the enterprise AI market is still moving too fast to bet everything on one vendor.

What actually happened

Zoph rejoined OpenAI in January 2026 after leaving Thinking Machines Lab, the startup he co-founded with Mira Murati, and was tasked with rebuilding the company’s enterprise sales operation. According to TechCrunch, Zoph has left that post and joined Google DeepMind as VP of Research, focused on reinforcement learning and post-training — an area he’d already worked in during his earlier years at Google Brain.

At the same time, Menlo Ventures’ mid-2026 report on LLM spend — one of the industry’s go-to benchmarks for where enterprise AI budgets actually go — confirms that Anthropic has overtaken OpenAI as the leading enterprise AI vendor, with Google in third place and gaining ground on the back of Gemini. None of the three has stood still: the same report notes that total spend on model APIs has more than doubled in six months.

Why it matters even if your company doesn’t buy directly from any of these giants

It’s tempting to write this off as Silicon Valley noise that doesn’t touch a Spanish SME. It does, for two specific reasons:

• Leadership changes hands faster than annual contracts do. If your company adopted AI a year ago by betting on “whichever vendor was winning at the time”, that same vendor may no longer be the strongest or the cheapest fit for your use case today.
• Turnover in key roles is a signal, not an anecdote. The person responsible for selling AI to enterprises lasting five months in the role, at the company that until recently dominated the market, says something about how unsettled this industry still is internally — even when it looks consolidated from the outside.

Signs your company is too dependent on a single AI vendor

• Prompts, agents or integrations are written directly against one model’s API, with no middle layer that would let you swap it out.
• Nobody on the team has compared the current vendor’s cost or quality against alternatives in the last six months.
• The contract locks you in for a long term, or volume discounts penalise leaving early.
• All the team’s know-how about “how AI works here” is tied to one vendor’s interface, not to the business problem it solves.

None of these on its own is a reason to panic, but two or more together are worth a review before you sign the next contract or expand the current one.

Staying flexible without complicating your operations

You don’t need to juggle multiple vendors to be protected — you just need to avoid building anything in code that’s tied to one specific model. A thin orchestration layer — LangChain or equivalent — lets you switch between GPT-5, Claude and Gemini depending on which one handles a given task better, without rewriting your business logic. It’s exactly why at Evicron we work indistinctly with GPT-5, Claude, Gemini, Llama or Mistral: the vendor gets picked based on what suits each project, not the other way round, and the code stays yours if you ever decide to switch.

How we approach this at Evicron

Our AI consulting always starts by evaluating which model fits a given use case best, not which brand happens to be trending that quarter. We covered what to ask before hiring outside help in more detail in our guide to AI consulting; the core idea here is simpler: in a market where the leader changes within months, the best protection for a company isn’t picking the right vendor — it’s not depending on any single one.

We’ve been building applied AI for companies across twelve sectors throughout Europe since 2019, with 200+ projects delivered and without tying ourselves — or you — to a single model maker. If you want to check whether your current AI setup leaves you room to manoeuvre or has locked you into one vendor, get in touch: the first session is free and we reply within 24 hours.