算力投資對利率變化不敏感

Frontier artificial intelligence labs have discovered that capital expenditures can outrun the discount rate.

Central bankers have started to price the artificial intelligence buildout as a macroeconomic shock. At the International Monetary Fund meetings in Thailand and recent central banking symposiums, Federal Reserve Chairman Kevin Warsh and regional officials pointed to tech capital expenditures as a persistent source of price pressures. With the industry projected to spend nearly one trillion dollars on hardware and facilities over the coming year, roughly three percent of domestic output, monetary authorities are confronting a structural mismatch between their primary policy tool and the capital cycle driving inflation.

The core difficulty lies in the interest-rate elasticity of compute. Traditional capital expenditure models assume that higher borrowing costs raise the hurdle rate for new projects, weeding out marginal corporate investments and slowing aggregate demand. Frontier artificial intelligence infrastructure does not obey this discount rate. Hyperscalers like Microsoft, Alphabet, and Amazon hold tens of billions of dollars in liquid balance sheet cash and generate immense operational cash flows from legacy monopolies. For venture-backed labs and frontier sponsors, capital allocation is governed by terminal race dynamics rather than near-term weighted average cost of capital. When management teams view cluster size as an existential hedge against obsolescence, a policy rate of four percent does not alter the buildout schedule.

The standard defense from monetary optimists rests on long-run total factor productivity. If advanced models automate cognitive labor, optimize power grids, and streamline administrative overhead, the supply side expands, generating secular disinflation over time. Yet for a central bank operating on an eighteen-month policy horizon, productivity gains function as long-dated call options with uncertain expiry dates, while datacenter consumption is a spot transaction. Transformers, high-voltage substations, backup generation capacity, and specialized mechanical contractors require immediate capital and physical resources. The demand shock hits local electrical utilities and supply chains today, whereas the deflationary dividends remain hypothetical future cash flows.

Because compute spending remains largely indifferent to policy rates, the transmission channel of monetary tightening lands entirely on the credit-sensitive periphery. When central banks maintain restrictive rates to counterbalance tech-driven price pressures, borrowing costs escalate for regional enterprises, commercial real estate developers, homebuilders, and levered households. A manufacturing firm financing a warehouse expansion or a regional bank managing commercial loan books must absorb the tightening. In effect, interest-rate-sensitive industries are forced to idle productive capacity so that sovereign balance sheets and cash-rich tech platforms can continue purchasing power generation and advanced packaging without triggering runaway headline inflation.

This dynamic extends directly into municipal utility rate design. Industrial datacenters require continuous, uninterrupted base-load power, prompting utilities across key regional grids to negotiate massive capital investments in transmission lines and generation capacity. Even when utilities structure power purchase agreements to protect residential ratepayers, the aggregate demand strains existing grid capacity, driving up wholesale spot power costs during peak periods. Regulators face a choice between allowing localized inflation in utility bills or approving rate increases that distribute the capital expenditure across small commercial customers. The monetary medicine fails to curb tech power consumption; it simply transfers the carrying cost to non-AI power users.

Monetary policy is designed to manage aggregate demand across an economy of price-sensitive actors. When the dominant investment wave of a decade operates outside that pricing discipline, interest rate adjustments cease to act as a universal brake. Instead, restrictive policy becomes a tool of selective resource reallocation. Small businesses and credit-dependent borrowers absorb higher financing charges, not because their own operations are overheating, but because their capital must be rationed to create fiscal headroom for an unhedged compute race.

Originally published at deanlee.info.