Good morning. It's Thursday, June 11, and we're covering the race toward self-improving AI, SpaceX's high-stakes AI infrastructure gamble, and why the future of AI may be built in space.

Plus: yesterday's Real or AI verdict, and a guest essay arguing that as AI gets smarter, the real bottleneck migrates to chips, energy, and manufacturing.

SpaceX's new AI factory is built to churn out thousands of orbital data centers, not just satellites. The company unveils its 11-million-square-foot Gigasat facility in Texas and targets 1 GW of space-based AI compute annually by late 2027. Each AI1 satellite carries 150 kW of compute, implying more than 6,000 launches per year, while Elon Musk projects 100 GW annually by 2030.

The biggest threat to SpaceX's IPO may come from passive investors who could be forced to buy in. Sen. Elizabeth Warren urges the SEC to delay the Nasdaq debut, citing the company's $1.75 trillion valuation, Elon Musk's voting control, and accelerated index inclusion. Her letter says major funds could absorb $15 billion to $30 billion in mandatory purchases despite SpaceX posting a $4.94 billion loss in 2025.

China-linked hackers are increasingly stealing AI technology rather than building it internally, according to CrowdStrike. The cybersecurity firm says Chinese entities account for more than 58% of state-backed cyberattacks targeting tech companies and their AI assets. Covering the year through March 31, the report links the campaign to efforts to narrow the U.S. technology gap amid chip export restrictions.

Devin Kim, a founding xAI post-training engineer, filed suit Tuesday in California state court against xAI and SpaceX, alleging he was fired for repeatedly raising concerns about Grok's potential to spread discrimination and information on weapons of mass destruction. Notably, the complaint does not target Elon Musk, who Kim's lawyers say had ordered xAI to follow the law and implement safety processes; blame falls instead on co-founder Jimmy Ba (who left xAI earlier this year) for allegedly ignoring those directives, telling Kim "AI will kill us all anyway."

Anthropic's Fable 5 (Mythos) stuns with unmatched coding, autonomous workflows, and massive AI agent orchestration, but at a premium price.

The following piece is by guest contributor, Ben Pouladian, CEO of BEP Holdings and publisher of BEP Research, where he covers AI, semiconductors, energy, and infrastructure.

We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.

That is the Anthropic Institute, writing about AI that helps build the next, better AI. And they put a hard number on it: as of May 2026, "more than 80% of the code we merge into Anthropic's codebase was authored by Claude." Two years ago that figure was a rounding error. Whatever you think about where this ends up, the first leg of it is not hypothetical. It is already running inside one of the best AI labs in the world. → Read the full article here.

A new wave of increasingly capable AI systems is reviving debate over “recursive self-improvement” (RSI), the idea that an AI could eventually design and build more advanced versions of itself with little or no human involvement. The article points to rapid progress at companies like Anthropic, where AI now writes most of the firm's code, alongside experiments showing models can already optimize aspects of AI development.

While some researchers warn RSI could accelerate progress beyond human oversight, others note practical constraints such as computing power, data availability, and specialized tasks that remain difficult to automate. The bigger takeaway is that AI-driven research may speed up dramatically even before fully autonomous self-improving systems become possible, raising urgent questions about governance and safety. → Read the full article here. (Paywall)

SpaceX is set to go public at a $75 billion offering, but its long-term valuation increasingly rests on Elon Musk's vision of orbital AI data centers rather than rockets alone. The strategy depends on achieving three difficult engineering milestones: fully reusable Starship launches, mass production of AI-computing satellites, and building a U.S. chip foundry to supply them.

Analysts at Morningstar and NYU's Aswath Damodaran value the company well below its proposed $1.8 trillion valuation, citing uncertainty around the AI business while praising the profitability of launch services and Starlink. → Read the full article here.

Musk rented the data center xAI abandoned to Google and Anthropic for $26B a year. Then priced SpaceX’s IPO the following week at $1.75 trillion. Genius business move, or a pump?

Sharing what everyone said tomorrow.

MIT spinout Ferveret is applying heat-transfer techniques from nuclear reactors to make AI data centers more efficient, using a water-free liquid cooling system that boils at the chip surface to remove heat. The company says its technology improves computational power efficiency by 15% over leading liquid-cooling methods and, when paired with its control software, can generate 35% more AI tokens from the same amount of electricity.

Ferveret is already testing the system with customers including CleanSpark, FuriosaAI, and Switch, while pursuing partnerships with major cloud providers. As AI drives a surge in data center construction, the startup is betting that reducing both energy and water consumption could ease one of the industry's biggest infrastructure bottlenecks. → Read the full article here.

Seattle Pauses New AI Datacenters: City approved a one-year ban on new AI-focused facilities to study energy use, land impacts and power costs.

Microsoft Limits Claude Fable Use: Employees can't access Anthropic's newest model internally as Microsoft reviews its data retention policies.

Palantir Helps Hospital Spot Sepsis Early: Tampa General says its AI monitoring system has saved 886 lives by catching sepsis sooner.

Google Cuts AI Plus Price: Google slashed its AI Plus plan to $4.99 and doubled storage, turning AI subscriptions into a price war.

Yesterday’s passage was AI. Tune in next week for another Real or AI challenge.

Thanks for reading. See you next time!
— Matthew Berman, Nick Wentz & the Forward Future Team