Why the Fed Is Watching AI Demand More Than Anything Else Right Now

Fed's New York President John Williams just named artificial intelligence as his main inflation concern, and he's not being vague about it. If AI-driven demand keeps pushing prices higher in a sustained way, the Fed's ready to hike rates again.

That's a shift. For the last year, the Fed's been watching services inflation, wage growth, and housing costs. Now Williams is saying AI demand could be the thing that forces their hand. He laid this out at a New York Fed event on Thursday, and the framing matters because Williams is vice chair of the FOMC, which means his view carries weight when the committee votes.

Here's the key line: "If this creates a sustained impulse to demand relative to supply in inflation, I do think that's the kind of situation where you don't look through this." Translation: if AI keeps pushing prices up, they're not ignoring it like they might with a temporary supply shock.

Williams gave a specific number to watch. If core PCE (the Fed's preferred inflation gauge) runs at 0.2% monthly for the second half of 2026, that keeps them on track for their 2% annual target. Anything higher than 0.2% per month and he's saying inflation's more persistent than they want.

That 0.2% monthly target is important because it's the line between "we're good" and "we need to tighten." It's not a forecast, it's a threshold. If we're sitting at 0.25% or 0.3% monthly by October, the probability of a rate hike in Q4 goes way up.

Right now, nine FOMC members already penciled in at least one quarter-point hike for 2026 in their June projections. The minutes from that meeting showed "a few" participants saw a case for hiking immediately. That's not consensus yet, but it's trending that direction.

AI spending isn't behaving like normal capex cycles. Companies are building data centers, buying chips, and locking in power contracts at a pace that's creating supply bottlenecks across multiple sectors. Energy infrastructure, semiconductors, cooling systems, real estate near fiber hubs, all getting hit with demand that wasn't in anyone's baseline six months ago.

When demand outpaces supply in a concentrated way like this, it doesn't just show up in tech stocks. It shows up in industrial metals, construction costs, power prices, and eventually wages in the sectors that support all that infrastructure. That's the "sustained impulse" Williams is talking about.

The Fed's model for this kind of demand shock is still catching up. They're running scenario analysis now, which the June minutes detailed. Williams called it a "collective reaction function," which is Fed-speak for "we're figuring out how to respond if X happens." The fact that they're doing this exercise publicly tells you they're serious about the scenario.

If you're trading indices or tech-heavy ETFs, the setup here is tricky. Rate hike expectations don't move in a straight line. They build gradually through data releases, then get priced in fast once the Fed signals clearly.

The $SPY and $QQQ both sold off in June when the Fed minutes first hinted at hike discussions, then bounced when the data came in softer than expected. That's normal. What's less normal is how little conviction there is in either direction right now. Volume's been declining on rallies and distribution days aren't showing panic, just rotation.

If core PCE prints above 0.2% monthly in the July or August reports, that's when rate hike odds start getting priced more aggressively. The bond market will move first, then equities follow. Watch the 2-year Treasury yield as the leading indicator for this. When that starts climbing on inflation data instead of growth concerns, it's a different trade.

Chairman Kevin Warsh launched task forces in June to review the Fed's models, including one specifically on productivity. That's relevant here because if AI's actually boosting productivity fast enough, it could offset the demand-side inflation pressure Williams is worried about.

Productivity gains let the economy handle more demand without overheating. If companies are getting more output per worker or per dollar of capex because of AI tools, that changes the math. The problem is productivity data lags by months and gets revised constantly, so the Fed's flying a bit blind on this part.

Warsh gave the task forces six months to report back, which puts the findings around December or January. If they conclude AI's raising the speed limit for growth without inflation, that's a dovish signal. If they conclude it's just demand without the productivity gains to match, that's hawkish.

The scenario where this gets messy is if AI demand stays hot but the broader economy slows. That puts the Fed in a bind. Do they hike into weakness to control inflation, or do they hold and let prices run hotter than target?

Historically, the Fed's chosen to control inflation even if it means a sharper slowdown. Volcker in the early '80s is the extreme example, but even post-2021, they hiked through signs of economic softness when inflation was the priority. If Williams is naming AI as his top concern, that tells you where his bias sits if he has to choose.

The other risk is measurement. Core PCE excludes food and energy, which means if AI demand is pushing up power costs or industrial commodity prices, it might not show up cleanly in the Fed's preferred gauge right away. By the time it does, they could be behind the curve.

Three things matter most over the next few months:

Core PCE monthly prints. The July report drops in late August. If it's 0.2% or under, the Fed stays patient. If it's 0.25% or higher, hike odds jump.

The 2-year Treasury yield. Currently sitting around where it was in early June. If it breaks above the June highs on inflation data, not on growth optimism, that's the market pricing in hikes.

AI capex guidance from big tech. Microsoft, Google, Amazon, and Meta all report in late July. If their capex guidance for Q3 and Q4 stays elevated or increases, that confirms the demand story Williams is concerned about. If it moderates, the inflation pressure eases.

The Fed's not guessing here. They're watching specific data and they've told us what the threshold is. If core PCE runs hot, they hike. If it doesn't, they don't. The only question is whether AI demand keeps pushing that number higher or if it stabilizes on its own.