The Frontier AI Race: Where Things Stand, March 2026

OpenAI just declared Code Red for the second time in as many months. This one isn't some form of a course correction, it's a sign of second warning on the same issue inside the same organisation which shows that the first one didn't land: the culture didn't shift, strategy didn't pivot enough, and the focus didn't sharpen. Fidji Simo, OpenAI's CEO of Applications, stood up in front of the entire company and said what the market (or at least Twitter has been talked about for months) already knew: Anthropic is eating them alive in enterprise.

The numbers confirm it:

The company that invented the category is now chasing the company that was supposed to be the safety-focused research lab. I've been calling it for a while that OpenAI has lost its default position as the market matures beyond vending machine chatbot experiences.

The root cause isn't a model quality problem, not even close. It's a focus problem, and it's been building for years.

Look at what OpenAI shipped (or tried to ship) in the last 12 months:

I'm probably missing announcements too. Every one of those is a resource drain that competes for engineering talent, leadership attention, and capital, and none of them connect into a coherent enterprise strategy.

Sora burned $15M per day in inference against $2.1M in lifetime revenue before it was killed; the Atlas browser never found an audience; and the hardware experiments haven't evidently gone anywhere (yet). When a company tries to win every disparate race simultaneously, it finishes in a mediocre position in all of them.

Then there's the Microsoft situation. The company that wrote the $13.8B cheque that made OpenAI who they are today is now threatening to sue them. Sam Altman handed a $50B cloud deal to Amazon instead of Azure, and Microsoft's response was fairly blunt: they know their contract, and they'll enforce it. On the same day, Microsoft restructured its entire AI organisation and freed Mustafa Suleyman to pursue superintelligence development in-house. Not two stories, but one: Microsoft is simultaneously fighting to protect the partnership and preparing to replace it.

Anthropic is the mirror image of all this. They picked one lane, enterprise, started niche by disrupting software engineering (their core domain) and went all in.

Every single week, Anthropic compounds on the same thesis. Revenue hit $19B annualised by March (80% enterprise), and 73% of companies buying AI for the first time are choosing Anthropic.

OpenAI's response to Code Red tells the same story in reverse: kill Sora, shelve the browser, redirect everything to Codex and enterprise. They're now trying to do what Anthropic has been doing for 18 months.

Codex quadrupling its weekly active users to 2M+ since January is the one bright spot. It's also the product most directly competing with Claude Code, which tells the story of exactly where they believe the recovery will come from, and where budgets are investing.

This isn't a two-horse race, though. It's three companies playing three different games.

Google is the one I find myself thinking about most. They arguably hold the strongest structural hand of anyone in AI: Chrome (the default gateway to the web), Android (dominant global smartphone share), Search, Workspace, YouTube, and now the AI layer inside Apple's Siri across 1.5 billion devices. Layer on the full compute stack (TPUs, cloud, energy infrastructure) and Google controls both where AI is accessed and what it runs on.

Gemini models keep improving, with Deep Think hitting 84.6% on ARC-AGI-2. Google Cloud revenue surged 48% to $17.7B quarterly. But the nuances is that having the strongest structural position and converting it into business dominance are not the same thing. Google doesn't need to win the frontier race. When capability gaps between Gemini, Claude, and GPT are measured in weeks rather than years, distribution beats innovation. They understood this with search in 2005 when Apple made Google the Safari default. The same playbook is running with Gemini now.

The challenge for Google is Microsoft remains the enterprise gatekeeper: the tenant, the licensing relationships, the CIO trust. What I'm increasingly seeing is a three-layer pattern:

Businesses are choosing Anthropic as a standalone product. Not just accessing Anthropic models through Microsoft's tenant, but buying Claude directly for their highest-value work. Google's risk is that surface dominance without enterprise conversion means becoming the plumbing, not the product, but the pipes. I don't think Google is winning right now (although they had a killer final 9-months in 2025 with the release of innovation that outpaced most), but I'd be very careful about betting against the company that turned search defaults into a $2T business. I see them reigning supreme in areas like creative with their innovation in Veo, Nano Banana, Stitch, Lyria, and more. But winning enterprise through the creative department is a stretch.

This feels like the early cloud wars all over again, except the stakes are 10x larger and the timeline is compressed into months instead of years.

Key takeaway for me is that the enterprise AI market will not be a one-horse race (B2B tends not to be a winner takes all, or most, category for most industries). Two Code Reds in 90 days tells the story of where the momentum sits, and where it doesn't.

I'll update this next month, or sooner if something big shifts. In this market, something big shifts about every two weeks.