Seagate is up 645% in 12 months. Hut 8 is up 432%. Both trade on the premise that AI infrastructure spending will not only continue at its current pace but accelerate, and that every dollar of the $405 billion in hyperscaler capex will convert cleanly into revenue for the companies downstream. That premise requires a level of execution precision that no capital cycle in modern history has delivered. These stocks are not a buy. They are a bet that nothing goes wrong, and the market is offering almost no compensation for the possibility that something does.
What the Price Already Knows
Broadcom's $8.4 billion AI revenue quarter is real. So is Micron's 25% gain in 12 weeks. I am not disputing the revenue. I am disputing what the market has done with it. Seagate trades at 56 times earnings. Hut 8, a company that spans Bitcoin mining, energy, and AI cloud platforms, carries a $6.54 billion market cap on a business model that did not exist in its current form 18 months ago. When you pay 56 times earnings, you are not buying today's results. You are buying 5 years of compounding growth with no interruption, no margin compression, and no capital misallocation. That is a very specific future.
Consider what a retail investor is actually buying at these levels. If you put $10,000 into Seagate at a 56 P/E, you are paying $178 for every dollar of current earnings. For that investment to return even 8% annually over 5 years, Seagate needs to roughly double its earnings while the multiple holds. If the multiple compresses to 30, which is still generous for a storage company, earnings need to nearly triple. That is the math behind the enthusiasm.
The last time infrastructure stocks posted triple-digit annual gains on the back of a massive corporate spending cycle was the late 1990s telecom buildout. JDS Uniphase, Nortel, Global Crossing. The capex was real then too. Carriers spent over $500 billion building fiber networks. The revenue arrived. The problem was that it arrived slower than the stocks had priced, and when utilization rates disappointed, the correction erased 80% to 95% of peak valuations. The fiber itself was fine. The investors were not.
The Assumption Nobody Stress-Tests
The bull case rests on a simple chain: AI needs compute, compute needs power, power needs infrastructure. Each link is true. But the chain has a weak joint that almost nobody discusses: monetization. The $405 billion in capex is being spent by 4 companies. Those 4 companies need to generate returns on that spending, which means their AI products need paying customers at scale. Amazon Web Services re-accelerated to 20% revenue growth, but AWS is a mature cloud business. The AI-specific revenue streams, the ones justifying gigawatt data centers, are still early.
If enterprise AI adoption follows the typical S-curve rather than the vertical line the market has priced, the hyperscalers will slow their capex growth. Not stop it. Slow it. And slowing is all it takes. When your stock is up 432% on acceleration, deceleration is a crash catalyst.
I will grant the bulls one thing: Constellation Energy's 20-year contracts with Microsoft and Meta are genuinely durable revenue. That is real visibility. But Constellation at 29.6 times forward earnings is already pricing in that durability, and the ASCE's D+ grade on U.S. energy infrastructure means the grid upgrades those contracts depend on face permitting delays, labor shortages, and cost overruns that no analyst model accounts for honestly.
Then there is the natural gas variable. Fidelity's own portfolio managers noted that abundant domestic supply could scale fast enough to cap energy prices. Cheaper gas is great for hyperscalers. It is terrible for the energy stocks trading at all-time highs on the assumption that power scarcity will persist. GE Vernova and BWX Technologies hitting records on grid strain makes sense only if the strain lasts. A supply surge turns their tailwind into a headwind.
The VIX is sitting near 14. Credit spreads are tight. Insider selling across semiconductor names hit a 3-year high last quarter. Every signal that usually precedes a repricing is flashing, and the consensus response is to point at Broadcom's revenue and keep buying. In 1999, the consensus pointed at Cisco's revenue. Cisco's revenue was real too. The stock still hasn't recovered its March 2000 high.
