The Two-Horse Race
Essay: Anthropic and OpenAI are the only AI companies that matter right now.
Over the last two months, the AI race has collapsed into a dogfight between two companies — Anthropic and OpenAI.
Both have taken convincing leads with frontier AI models — Anthropic just broke through $30B ARR (up from $9B just three months ago!) and OpenAI’s Codex coding model is breaking benchmarks.
The next six months will either make or break these companies, but one thing is certain — for now, it’s their race to lose and the window for Google, Meta, Microsoft or Apple to do anything about that is closing fast.
Falling Behind
When was the last time you used one of Meta’s AI models?
Exactly.
Despite spending upwards of $30B+ in the last year to acquire top AI talent, data centers and GPUs, Meta has failed to release a single model in 12+ months. Llama is dead, with multiple open source Chinese models crushing them. The latest rumors indicate Meta might even be exploring ditching the model training entirely and licensing Google’s Gemini.
Which reminds me… Where is Google?
After releasing Gemini 3 last year (and blowing everyone’s minds) Google has slowed down model releases with their latest 3.1 release failing to live up to expectations. The model frequently hallucinates and does not code or reason as well as Opus or GPT.
So that leaves us with Microsoft who is entirely dependent on ChatGPT (it owns 27% of OpenAI) and Apple who is paying Google $1B annually to license Gemini. The hilarious part is that Apple wanted to use Anthropic’s Claude but Apple didn’t want to pay up. 😆
Finally we have… Grok.
The mass exodus from xAI over the last month has done little to calm my nerves that Elon has dropped the ball here BUT at the same time they are currently in control of the world’s largest GPU cluster and most valuable social media data moats with X. They have a chance to catch up but OpenAI and Anthropic are seemingly running further away with it every day.
The Coding War
It’s pretty obvious to anyone paying attention what really matters…
The model that wins coding wins everything.
Coding models generate the most revenue, attract more developers, and can be used to build the next versions of themselves.
Claude Opus and ChatGPT Codex are by far the best models available. Winning AI coding means you can assume any software challenge of the past, present or future can and will be solved.
The numbers back all of this up, Codex weekly users are at 2M (up 5X since start of the year) and Anthropic’s $30B ARR is highly dependent on Claude Code.
The Next Models Will Be AGI
As I’m typing this, Anthropic just revealed their next model Claude Mythos has discovered 1000+ security vulnerabilities across the top software systems with 99% of them going unfixed… that means if they were to release their new model today — a malicious user could hack into the top operating systems used by companies and users globally.
Dystopian cybersecurity concerns are no longer a thing of science fiction, this is 100% real and Anthropic is refusing to publicly release the model until fixes have been put into place. Say what you will but if the model is capable of executing hacks most elite security researchers aren’t capable of defending themselves, then we’ve reached a point where AI has surpassed human intelligence aka AGI.
OpenAI is not too far behind either — their upcoming model code-named ‘Spud’ is the “result of 2 years worth of research” according to president Greg Brockman. Sam Altman has claimed in multiple interviews recently that they believe AGI will be achieved internally by end of year and he even dropped a manifesto laying out proposed economic reforms needed to live in a post-AGI world.
Now take into account that not only will these models get smarter than humans but also build themselves — this means that we will get: More innovation, faster than ever before with near-zero mistakes.
Never before have we seen technology advance in such a hyper-exponential way as this, and OpenAI and Anthropic are leading here.
It’s going to be hard to keep up if you don’t drop a similar model within the next few months.
Pressure Makes… Dynamite
I wish I could claim the startups are keeping up with the pressure but the cracks are definitely appearing.
In the last couple weeks, Altman has shut down Sora and any other focus that isn’t explicitly on improving ChatGPT and coding models. The pressure of keeping up with Anthropic has forced him to abandon what made OpenAI arguably special — retail AI apps.
It’s now a dogfight to the finish.
Dario on the other hand has had a rough couple weeks following Claude Code’s source code leak, and it’s ironic that Anthropic’s security practices seem to be so sloppy even as it leads the charge against the biggest cybersecurity model threat since the start of the AI rush.
The question is… will these two companies be able to keep it all up?
The model releases are now happening every month (it used to be every 90 days), the compute demand is getting more and more expensive, the token demand is growing higher and higher, and in some cases they can’t afford to deliver on everything.
OpenAI and Anthropic are in a very precarious position. They have everything to gain and lose simultaneously.
It’s going to be a very important 90 days going forward. We will see the two most powerful models launch ahead of some of the world’s biggest IPOs.
Clock is ticking.
Thanks for joining us for our latest issue. Now go listen to our podcast :)