Deep Dive
AI buildout is slowing in politics, not reality
Forecasting experts keep missing AI capability gains and adoption curves, consistently underestimating both model performance and revenue growth from major labs. The narrative around data center bans sounds catastrophic — 300 US towns have paused new facilities and several states threaten restrictions. But the actual impact is minimal. Semi Analysis data shows only 1.5 gigawatts of the 38-gigawatt pipeline has been delayed, meaning 96% of approved power is proceeding unaffected. The political loudness around AI safety and moratoriums creates fear, but actionable delays are a rounding error. Politicians make threats that can't be enforced and will only drive companies out of restrictive states, similar to how billionaires already fled California. The real constraint isn't regulation — it's getting permits for infrastructure like power lines into new data centers, as Oracle discovered with Project Jupiter in New Mexico.
Space solves Earth's data center constraints
Because terrestrial power and cooling infrastructure create bottlenecks, the logical next step is orbital compute. Google is launching Project Suncatcher, a prototype satellite equipped with TPUs, on a Falcon 9 Transporter mission within days. An Nvidia Vera Rubin box already orbits, testing payload physics in space. Critics called space data centers impossible — claiming no cooling, heating, electricity, or connectivity exists in orbit. But those obstacles are already solved. Starlink's 11,000 satellites provide global connectivity. Space itself offers free cooling via thermal radiation. The real insight is that SpaceX controls the entire supply chain. They launch via Falcon 9, connect via Starlink, and will scale via Starship when Google, OpenAI, Anthropic, and others need to deploy more compute off-world. No competitor can replicate this vertically integrated moat. When the compute hunger grows beyond Earth's power grid capacity, there's only one door to knock on.
Meta's Muse and AI agents unlock crypto payments
Meta's stumbling with AI finally paid off. Muse, a personal assistant for social apps that handles tasks like sending messages or booking dinners, hit 300,000 downloads in its first phase — more than WhatsApp, Messenger, and Threads combined. Alexander Wang, Meta's AI head whom Zuckerberg hired for a reported $5 billion package, created value that's already added $200 billion to Meta's market cap. The hire looked absurd publicly, but the ROI speaks. More importantly, Black Rock confirmed that AI agents will fuel crypto adoption because agents need 24/7 micropayments, can't open bank accounts, and operate too fast for legacy banking rails. Everything routes through stablecoins and crypto payments. Solana is already winning this space and hired institutional veterans like Rachel from Binance as chief sales officer and Jamal from Polygon as GM of payments to bridge TradFi into tokenization. This is synergy: AI agents demand crypto rails, and Solana is positioning itself as the native money layer.
Memory shortage and GPU economics make scarcity profitable
Memory is sold out into 2027 with prices soaring beyond affordability. Customers ordering $10 billion worth of memory are lucky to receive $50 billion. Supply contracts remain massively unfulfilled because few companies planned for the chip crunch. Elon Musk publicly thanked Micron for actually fulfilling his orders while other players got nothing, proving that scale and foresight matter. The immediate play is memory stocks — the risk-reward ratio is insane through 2028. Separately, Nvidia B200 GPUs are generating $7.88 per hour in rental revenue, meaning they pay for themselves in roughly 50,000 hours or about 2 years of utilization. That perpetual income stream persists even as newer architectures like Vera Rubin launch. Old hardware stays economically viable because demand vastly exceeds supply. This creates a 10-year tail of scarcity-driven profitability in chips and infrastructure.
Stargate financing is cracking under complexity and junk spreads
The $500 billion Stargate project announced by OpenAI, SoftBank, Oracle, and others looked ambitious but is already showing strain. Oracle filed a force majeure on Project Jupiter, the New Mexico data center supplying power to Stargate, because permits for the gas pipeline never materialized. When permits slip on critical infrastructure, the entire cascade breaks. SoftBank priced $11.1 billion in senior notes at junk levels. Oracle's credit default swaps hit record highs with debt trading at 90 cents on the dollar — a haircut signal. Core Weave, the neo-cloud player financing the stack, is raising additional debt at dilutive terms. The complexity of multiple chefs in the kitchen — SoftBank, OpenAI, Oracle, Blue Owl, Bloom Energy — creates coordination risk. Any single permit delay or financing hiccup ripples through all parties. This isn't because the companies are bad or the thesis is wrong. It's because infrastructure financing at this scale hits regulatory, permitting, and coordination headwinds. Watch this through 2028 as a potential house of cards.