xAI is looking more like a datacentre REIT than a frontier lab

That framing mostly held up. People broadly accepted that xAI and SpaceX stumbled into a very valuable position by controlling scarce GPUs, power, and datacenter capacity at a moment when the rest of the market is short. Several commenters pushed the point further: this is not really a REIT analogy at all, because REITs lease space and power, while xAI is selling live AI compute. In today’s market that sits higher up the value chain and can earn extraordinary margins. The stronger version of the bullish case was simple: xAI may be losing the model race, but owning ready-to-run AI infrastructure is still a great business while supply is constrained. The main skeptical line was about duration, not present demand. Commenters kept coming back to the same tension. GPUs are usually depreciating assets, but current shortages have made even older hardware appreciate or hold value far longer than expected. That does not make them permanent stores of value. It means the market is in an unusual window where old H100s and even A100s still clear at high prices because power, permits, memory, and deployment capacity are bottlenecks. Many readers saw that as enough to justify short-term rental economics, but not enough to justify trillion-dollar narratives built on them. The other big thread was financial structure. A lot of people were uneasy about overlapping ownership, related-party incentives, and revenue that can make one company look healthier while also marking up another company’s stake. The sharpest version of that concern was not that every deal is literally fraudulent or circular financing in the accounting sense. It was that the AI stack is now packed with incentives to use expensive compute contracts, equity marks, and IPO timing to support valuations that public markets may eventually inherit. Others pushed back that this specific Google-xAI deal still looks like a straightforward response to a compute shortage. If you need capacity now and xAI has it, renting it is rational even if the cap table around it is messy. A second practical thread cut against the article’s tone. xAI’s speed in building Colossus impressed people, but many said the speed came from brute force, temporary gas turbines, and regulatory corner-cutting rather than some novel datacenter genius. That matters because it suggests the moat is not uniquely better execution so much as unusual willingness to absorb legal, environmental, and reputational risk. Several readers thought that also explains why rivals are willing to rent from xAI anyway. Compute scarcity is severe enough that brand risk and operational awkwardness are being tolerated. By the end, the center of gravity was clear. The market is treating frontier models and compute capacity as one big AI story, but they are not the same business. xAI may have a profitable infrastructure asset in the middle of a shortage. That can be true at the same time as the article’s broader point: if the revenue is coming from renting out GPUs, investors should stop valuing it like a company that is going to win on differentiated model intelligence.