Saturday, May 23. Latest on the Edge of AI: three of the most-watched private companies in the world are eyeing the same public market, a new Luddite movement is organizing against AI, and a Chinese lab is raising billions to chase AGI before profit. Let's get into it.
SpaceX, OpenAI, and Anthropic are all angling for IPOs in what's shaping up to be a test of how much public-market capital the AI boom can absorb [2]. Elon Musk's SpaceX, Sam Altman's OpenAI, and Dario Amodei's Anthropic are each in active discussions with underwriters. The Financial Times reports the three are competing for Wall Street's deepest pools of capital, and the timing matters: SpaceX is coming off a scrubbed Starship launch, OpenAI is hunting a $500 billion valuation, and Anthropic just closed a $30 billion round. The signal: the IPO window for frontier-tech companies has never been wider, but each of these three has a different risk profile — SpaceX has hardware risk, OpenAI has burn-rate risk, and Anthropic has the most concentrated revenue base. The market will have to decide which one gets the premium.
Different beat, different risk vector. Anthropic, the lab behind Claude, just warned that its Mythos Preview model is finding vulnerabilities faster than developers can patch them [1]. Working with about 50 partners under Project Glasswing, Mythos has uncovered over 10,000 critical vulnerabilities in system-critical software. The bugs are piling up faster than anyone can fix them. Anthropic says no company, itself included, has built safeguards strong enough to prevent misuse of these models. The Decoder reports this creates a high-risk transition period. What this changes: the conventional wisdom that AI vulnerability discovery is a net good. It is, until the discovery rate outstrips the remediation rate. My read: this is the first time a frontier lab has publicly admitted its own tool creates a liability faster than the industry can absorb it. That's a warning worth taking seriously.
Now, on the Chinese side. DeepSeek, the Chinese open-source upstart, is raising roughly $10 billion at a $45 billion valuation [14]. Founder Liang Wenfeng is telling investors he's prioritizing AGI research over short-term profits. The Decoder reports DeepSeek's message to backers is that they're not optimizing for quick returns. The angle: DeepSeek is the first Chinese lab to explicitly tell the market "we're going long on AGI, not short on revenue." That's a bet that either forces every other Chinese lab to match the AGI-first framing or gets undercut by a faster-moving competitor. Either way, the $10 billion figure is a floor, not a ceiling.
Last beat. Quietly, an open-source project called Spice is trying to solve a problem nobody's really solved yet [4] [unverified]. Spice is a decision layer that sits above AI agents — instead of the agent deciding what to do based on a prompt, Spice acts as a brain that knows your context, priorities, and constraints before execution. The Reddit post from the team frames it as the missing layer between "good at doing stuff" and "good at deciding what to do." Which matters because: every agentic system today — Claude Code, Codex, Hermes — has the same weakness. They execute well but decide poorly. Spice is open-source, which means it could become the default decision layer for the agent ecosystem if it catches on.
Three labs, three bets, one open-source fix for a problem none of the frontier labs have solved yet.
That is the edge for today.