#ai-infrastructure

SK Hynix's Nasdaq debut was the largest US listing ever by a foreign company. It controls roughly 60% of high-bandwidth memory, which is the actual constraint on AI compute. The harder question is whether memory has really stopped being a cyclical business, and I don't think it has.

Your support bot does not need a PhD. Most production AI work is repetitive, narrow, and latency-sensitive, which is exactly where a 3B to 14B parameter model wins on cost, speed, and privacy. NVIDIA Research now argues the same thing in a paper, and the pricing backs it up.

Nvidia was the obvious AI trade. The non-obvious one is whoever plugs the grid into the building. Hyperscalers are guiding to $725 billion of 2026 capex, and the binding constraint is no longer chips. It's turbines, transformers, and interconnection queues.

Real estate people ignored the AI trade. Tech people ignored REITs. In the middle sit Equinix and Digital Realty, landlords for the compute that runs GPT and Gemini. In 2026 they are up 36% while the rest of real estate barely moved.

Microsoft is paying to restart Three Mile Island. That is not a meme, it is a thesis. Here's how the nuclear trade actually splits between cash-flowing utilities like Constellation and pre-revenue SMR bets like Oklo and NuScale.

The warehouse trade is over. Logistics rents fell 4.5% in 2025 while data center rents hit a record $196.25 per kW/month at 1.4% vacancy. The next industrial asset class is here, and its scarce input is not land, it's power.

Anthropic went from $1B to roughly $30B in annual run-rate revenue in 15 months, and Claude now breaks most mornings between 9 AM and noon Pacific. The outages aren't random. They're concentrated exactly where Claude won: long-running coding agents, on the messiest multi-cloud stack in the industry, metered by a rate limiter that sometimes drains a weekly quota in 19 minutes.