AI Daily Briefing — Thursday, May 21st, 2026
Episode: General AI News
━━━━━━━━━━━━━━━━━━━━━━━━━

Good morning. This is your AI Daily Briefing for Thursday, May 21st. Three stories dominating the landscape: AI proves a math theorem humans couldn't crack for 80 years, OpenAI is about to go public, and NVIDIA just posted numbers that redefine what "big tech earnings" even means.

Let's start with the science headline. OpenAI announced that one of its models disproved a central conjecture in discrete geometry — the Erdős unit distance problem, posed in 1946. This is the first time an AI system has solved a major open problem in pure mathematics that resisted decades of human effort. It's not generating text or code; it's generating genuinely new mathematical insight. The paper hit number one on Hacker News with nearly 900 points and over 600 comments. This is the kind of milestone that changes how we think about AI's role in scientific discovery.

Meanwhile, the business side of OpenAI is making its own headlines. CNBC reports the company plans to confidentially file for an IPO as soon as tomorrow. After years of speculation about their unusual capped-profit structure, the public markets are about to get a piece of the company that started the generative AI wave. The filing comes at a moment when AI infrastructure spending is at fever pitch — more on that in a moment.

Speaking of talent moves: Andrej Karpathy, the former Tesla Autopilot and OpenAI founding member, has joined Anthropic. This is a major signal. Karpathy is one of the most respected figures in practical AI — his decision to join Anthropic over returning to OpenAI or going independent says a lot about where the center of gravity is shifting. Anthropic is also making infrastructure moves: they're expanding their deal with SpaceX for access to the Colossus 2 supercomputer, reportedly worth nearly 45 billion dollars over three years. The labs aren't just competing on models anymore — they're competing on who can secure the most compute.

And that brings us to NVIDIA. The company reported Q1 fiscal 2027 earnings yesterday, and the numbers are staggering: 81.6 billion dollars in revenue, up 85 percent year over year. Data center revenue alone hit 75.2 billion — that's a 92 percent jump. Jensen Huang said demand has gone, quote, "parabolic" driven by agentic AI. NVIDIA also raised its quarterly dividend 25-fold and authorized another 80 billion dollars in buybacks. The Vera Rubin architecture is on track for the second half of 2026. If anyone was wondering whether the AI infrastructure buildout is slowing down — it's not.

In the "AI is coming for jobs" category, two data points. JPMorgan CEO Jamie Dimon publicly stated AI will reduce jobs at the bank, even as they hire more AI specialists. And Intuit, the company behind TurboTax and QuickBooks, announced it's laying off 3,000 employees — 17 percent of its workforce — explicitly to refocus on AI. These are not small experiments. These are boardroom decisions at two of the most established companies in finance and software.

A few quick hits before we close. Figma launched a product design AI agent that can generate and edit designs. LinkedIn is cracking down on AI-generated comments, limiting visibility of content it deems low-effort or automated. And the first criminal charges were unsealed under the Take It Down Act against two men who allegedly posted thousands of nonconsensual AI deepfakes — the law's criminal provisions have been active for a year, but platform removal obligations just kicked in.

That's your briefing for May 21st. AI is proving theorems, filing to go public, and reshaping workforces — all before most people have had their second coffee. I'll be back tomorrow. This has been your AI Daily Briefing.