InvestAnswers: Oracle hits force majeure on Stargate's Project Jupiter as AI credit cracks show

Cracks appear in the AI financing stack even as forecasters keep underestimating AI's pace

James opens by citing Forecasting Research Institute data showing domain experts and even super-forecasters keep underestimating how fast AI models clear benchmarks and how fast labs generate revenue. He argues "the smart money keeps on making money and the experts are missing the mark." He also references his own Substack piece on AI agents killing "inertia businesses" that profit from customer laziness or confusion, such as subscription services and fee-heavy banks.

The bigger story is stress in AI infrastructure financing. Oracle sent a force majeure notice on Project Jupiter, a 2.5 gigawatt New Mexico data center tied to the half-trillion-dollar Stargate plan with SoftBank, after failing to get gas-line permits approved in time for a planned 2028 go-live. James notes Oracle's credit default swaps hit a record high and its debt trades around 90 cents on the dollar, which he calls junk level. SoftBank priced roughly $11.1 billion in senior notes, also junk-rated, tied to the same OpenAI/Stargate financing, and CoreWeave has raised additional debt that James says will dilute shareholders. He frames this as "one big house of cards" and urges caution, while stressing the permitting complexity, not bad companies or bad financing, is the core problem.

Separately, he cites semi-analysis data that only about 1.5 gigawatts of the 38 gigawatts planned across the US have been delayed by roughly 300 towns pausing new data centers, calling the political noise around bans "a nothing burger" relative to the scale of the buildout.

Meta's Muse becomes a surprise hit, memory shortages deepen, and Solana recruits for institutional adoption

James highlights Meta's new AI assistant, Muse, which he says has reached 300,000 downloads, more than WhatsApp, Facebook or Messenger by his account, despite Meta's prior AI missteps. He credits Alexander Wang, whom he says was mocked for a reported $5 billion Meta signing package, with adding roughly $200 billion in market value under his watch: "the question is not how much talent costs but how much value they can create."

On hardware, James says memory chips are "sold out into 2027," with customers lucky to get half of what they order, and points to Elon Musk having publicly thanked Micron for fulfilling supply. He also cites older Nvidia B200 GPUs generating about $7.88 per GPU hour, arguing scarcity, not obsolescence, is the dominant force in AI infrastructure economics right now.

On crypto's convergence with AI, James cites BlackRock's view that AI agents will need 24/7 micropayments that banks cannot support, pushing activity onto crypto rails. He says Solana is responding by hiring institutional-adoption veterans, naming a chief sales officer he identifies as Rachel from Binance and a GM of payments he identifies as Jamal from Polygon, calling the move "incredibly intelligent."

Google tests data centers in space via SpaceX, and James dismisses Bernie Sanders' push to restrict AI

James describes Google's Project Suncatcher, a prototype satellite carrying TPUs set to launch on a Falcon 9 Transporter mission he says is about six days out, as the first real test of data centers in orbit. He argues SpaceX is the ultimate winner regardless of who launches first, since Starlink already provides connectivity and Starship will be needed to scale space-based compute: "There's only one game in town that can do that." He also notes SpaceX's Grok-based customer service bot resolves tickets for about 20 cents each versus a typical $28 to $65 per call, saving the need to hire roughly 200 staff.

Turning to politics, James criticizes what he refers to as Bernie Sanders' push to restrict "super intelligence," including a proposed AI kill switch and prison terms of 20 years for developers, calling it "absolute madness" and unenforceable. He argues that such bans would simply push AI companies out of restrictive states and that the buildout described by the Wall Street Journal as the largest in US history cannot be stopped, framing it as essential to future economic growth. He contrasts this with Singapore's plan to train 80,000 financial-services workers on AI.

James closes with a health anecdote reported in the New York Times about a hospice patient with a rare blood disease whose case, taken to the group Every Cure, led to a repurposed drug combination that reportedly aided his recovery within a week, which he offers as evidence of AI's potential in medicine. He also flags a joke about Meta's Muse requesting "full root access" to a user's computer as a genuine security warning, urging caution before granting AI tools access to banking or email accounts.