Fed's Lisa Cook ranks AI's mostly unspent $2 trillion buildout among her top 2027 inflation risks

The $2 trillion counts promises. Only a minor portion of it has gone out the door, according to Crypto Briefing's account of the Oakland speech ^[6], so most of the demand those plans put on electricity and other constrained inputs is still ahead. Cook's worry, as she put it to Williams, is the order of events: "I worry about when the productivity gains that would produce disinflation will come, and where the supply bottlenecks are going to be next" ^[4].

Inputs to the buildout have risen about 5% in a year ^[7], and the Fed's targeted measure ran at 3.4% in August ^[10]. The baskets differ, so the comparison is rough. Still, it puts AI's input costs about 1.6 percentage points ahead of the inflation rate the Fed targets ^[15]. That rate is itself 1.4 points over the 2% goal ^[16], and it has been above the goal for more than five and a half years ^[10].

Cook described the doctrine that would usually let the Fed ignore a cost shock of this kind. "Our conventional view used to be that we would look through the supply shocks because tighter monetary policy is not going to have an effect on oil prices, not going to have an effect on a war, but it could slow down employment and output outcomes that we would be more concerned about," she said ^[11]. Supply shocks have since had surprisingly persistent effects, she said, and the "optimal response" could now depend on the sectors affected ^[8][12].

So the Fed is not going to raise rates to slow construction; Cook said its tools are too blunt for that ^[5]. Her path from data centers to the policy rate runs through persistence instead. "The AI build out is potentially creating inflationary pressures that may not resolve very quickly," she told Williams ^[3].

The case for tighter policy can fail on timing alone. The announced spending could arrive slower than planned and ease the bottlenecks with no Fed action. Productivity could show up before the few years Cook allows for ^[1]. Or the next bottleneck could come from elsewhere: the CNA report notes the Middle East conflict could also crimp supply chains ^[14], and a 2027 price shock from oil would not be AI's. Neither report ties last month's quarter-point increase to AI ^[9].

I think the evidence makes AI capex a reason for the Fed to keep policy tight into 2027, or rather a reason not to look through the buildout's costs. It is not yet a reason to raise rates again. The case is wrong if buildout input costs slow from that 5% pace while the $2 trillion is actually being spent ^[7][6].