The Case for and Against the Capex Supercycle

Watch the depreciation schedules, not just the capex line. Hyperscalers are quietly extending the useful life they assign to servers, which flatters near-term margins even as the underlying hardware ages faster in AI workloads than the accounting assumes. When those schedules snap back to reality, the earnings hit arrives all at once.

Reading an AI earnings call is an exercise in separating booked revenue from backlog from ambition. Signed contracts and committed capacity are real; framework agreements and letters of intent are options on the future. The market routinely conflates the two, and that is where the mispricings live.

Hyperscaler capital expenditure guidance is the single most useful signal we get each quarter.

When a lab ships a new frontier model, the interesting question is rarely whether the benchmark went up. It is whether the price-performance curve shifted enough to unlock a category of application that was previously uneconomical. Watch the pricing page, not the leaderboard.

It is worth remembering that the enterprises paying for all of this are not buying models — they are buying outcomes. The vendor that can credibly tie its product to a line item a CFO already understands will win a budget fight against a dozen technically superior tools that cannot.

Open-weight models keep closing the distance to the closed frontier, and each release compresses the premium that proprietary providers can charge. That does not erase the moat — the frontier still leads on the hardest tasks — but it caps how much of the market the leaders can defend at the low and middle tiers.

The bull case for the capex supercycle rests on durable demand and expanding use cases; the bear case rests on the possibility that a great deal of this spending is defensive, undertaken because no incumbent can afford to be the one that under-invested. Both can be true at once, and the timing of the reckoning is the whole game.

Advanced packaging is the constraint hiding behind the constraint. Even when a foundry can print the logic, the number of chips that ship is gated by the capacity to bond memory stacks onto the compute die — and that line is booked out quarters in advance. The packaging vendors quietly set the ceiling on how fast supply can grow.

The interesting tell in a model launch is what the provider chooses not to charge for. Free tiers, aggressive rate limits, and bundled inference are competitive weapons aimed at locking in developers before switching costs exist. The pricing is a strategy document, and the giveaways say more about the roadmap than the benchmarks do.