AI Breakthroughs: OpenAI Cracks Legendary Math Problem, Anthropic Ups the Ante

An OpenAI model just disproved an eighty-year-old mathematical conjecture. That’s not a simulation or a game — a general-purpose reasoning model produced an original mathematical proof for the Erdős unit distance hypothesis, a problem that has been famously open for decades. Last week, we talked about autonomous AI workers transforming Wall Street; today, those same classes of models are starting to do original science. And that’s not the only ground that shifted. Anthropic is now paying SpaceX’s xAI one-point-two-five billion dollars a month for compute.

That’s not a typo. The deal, revealed in SpaceX’s IPO filing, gives Anthropic exclusive access to the Colossus 1 data center — a private city of over two hundred and twenty thousand Nvidia GPUs. This massive infrastructure is already paying off. Anthropic just reported its first-ever profitable quarter, projecting nearly eleven billion dollars in revenue for Q2 2026. The driver is runaway enterprise adoption of Claude Code, which now has a revenue run-rate over two-point-five billion dollars. And while Anthropic builds its empire, OpenAI’s CEO Sam Altman is signaling their next move… focusing on what he calls "personal AGI," designed to accelerate the goals of individuals, not just companies.

Let’s go back to that math problem. The Erdős unit distance hypothesis is a classic question in geometry. For eighty years, the world’s best mathematicians have tried to solve it. OpenAI’s model didn’t just find a solution — it found one in a way no human had seriously considered. Instead of using classical geometry, it used advanced algebraic number theory. It built a novel counterexample. This is the detail that matters. The AI didn't just brute-force a known path faster. It charted a new one. And this isn't just OpenAI marking its own homework. They sent the proof to external mathematicians, including Fields Medalist Timothy Gowers.

He called the result a “milestone for AI in mathematics.” This isn’t a tool that helps a mathematician work. This is a collaborator that can generate original ideas. It’s a co-discoverer. The immediate threat here isn't to mathematicians' jobs. The threat is to every research and development budget on the planet that assumes discovery only moves at the speed of human insight. That assumption is now obsolete. The new question for every lab isn't "what are our researchers working on?" It's "what is our AI working on while our researchers sleep?" Now let's talk about the money.

Anthropic’s one-point-two-five billion dollar monthly payment to xAI is one of the clearest price signals we’ve ever gotten for what it costs to compete at the frontier. This isn't a cloud contract. This is a utility-scale infrastructure commitment. Anthropic isn't renting servers; they've effectively bought the exclusive output of a power plant. A power plant made of two hundred and twenty thousand GPUs. This move fundamentally changes the economics of the AI industry. It’s why Anthropic can project ten-point-nine billion dollars in revenue and post a profit.

Their cost per dollar of revenue is dropping, from seventy-one cents to fifty-six cents, because they've locked in compute at a fixed, massive scale. They’re not subject to the whims of spot pricing on public clouds. They have certainty. Who does this threaten? Everyone who isn’t doing this. It threatens smaller AI labs who can no longer hope to compete by just raising a few hundred million dollars. The table stakes just went up to tens of billions. It also puts the big cloud providers on notice. Amazon, Google, Microsoft… they’re used to being the landlords.

But Anthropic just cut a deal directly with the owner of the power station. It signals a shift from renting capacity to owning it. So you have two things happening at once. OpenAI is demonstrating that AI can be a partner in pure scientific discovery. And Anthropic is demonstrating that you can build a massively profitable, industrial-scale business on the back of it. One company is creating knowledge, the other is creating the economic engine to power it. This is the new reality. AI is no longer just a software product. It’s an industrial utility, as fundamental as electricity.

And the companies that control the generation of that utility are the ones writing the rules for everyone else.