Wall Street Blinks, Washington Stalls

Two stories from this morning. They look unrelated. They aren't.

One is about money getting nervous. The other is about policy not moving. Both tell you something about the gap between how fast the AI buildout is happening and how slowly the systems around it are catching up.

Wall Street finally asks the question my clients have been asking

CNN ran a piece this morning titled "Big Tech's massive spending is back in focus on Wall Street." The framing is that after a few years of the AI narrative carrying the market, investors are starting to look hard at whether the capex actually pays back. Not someday. On a timeline a public-markets analyst can model.

This isn't new in private. It's new in public. I've sat in a lot of enterprise-IT conversations in the last twelve months where the CFO asks the same question the buy-side analyst is now asking out loud: where is the return, in dollars, attributable to this specific AI investment, in this specific quarter. The honest answer in most cases is "not yet, and we're not sure how to attribute it cleanly even when it shows up."

What I keep coming back to is the asymmetry. The hyperscalers are spending tens of billions a quarter on infrastructure that gets depreciated over a few years. The customers buying that capacity are mostly running pilots and a small number of production workloads. The math works if the production workloads scale fast. It doesn't work if they don't.

My read is that 2026 is the year the patience window closes for a chunk of the buyers. Not the believers. The middle. The ones who said yes to the pilot because the board wanted an AI line on the slide. Those budgets will get reviewed.

The NYT says AI is a national security risk

The New York Times opinion page ran a piece this morning arguing that AI is a national security risk and that we aren't doing nearly enough about it. The author's case is that there's bipartisan appetite to act, and that both parties just have to move faster.

I agree with the diagnosis. I'm a skeptic on the optimism.

The diagnosis is right because the surface area is enormous and growing. Model weights as exfiltration targets. Inference infrastructure sitting in jurisdictions with messy export regimes. Procurement pipelines for frontier compute that route through places nobody on a Senate Intelligence Committee would draw on a whiteboard. The boring operational version of "AI as national security" is mostly supply chain and access control, and we're not great at either.

Where I push back is the framing that bipartisan agreement is the bottleneck. It usually isn't. The bottleneck is that the clock speed of the technology is months and the clock speed of legislation is years, and even sympathetic policymakers are working with staffs that are still trying to define terms. Op-eds calling for action presume the action is teed up and waiting. Mostly it isn't.

What I tell clients in regulated sectors is to stop waiting for the policy environment to clarify. It won't, on a timeline that matters to your next renewal cycle. Build the controls now, document them, and treat the eventual regulation as something you'll comply with by adjustment, not by retrofit.

The thread

Read the two pieces back to back and you see the same gap from two angles. Wall Street operates on quarters. Washington operates on multi-year cycles. The AI buildout is happening on a timeline that's faster than the first and slower than the second. Capex commitments are getting locked in before the returns are legible, and policy frameworks are getting drafted after the operational risks are already real.

The enterprise buyers I work with sit in the middle of that gap. They're being asked by their CFOs to justify spend on a quarterly clock and by their risk and compliance teams to anticipate rules that don't exist yet. Both pressures are legitimate. Neither is going to relax.

I'd watch the next two earnings cycles closely. If a hyperscaler lowers AI capex guidance even modestly, that's the signal that the Wall Street pressure has crossed from commentary into board-level decisions. And if any meaningful federal AI legislation actually moves before year end, I'll owe the NYT op-ed page an apology.