The AI Reckoning Begins—And Powell Just Made It Worse

The past 72 hours crystallized something markets have been dancing around since November: the emperor has no clothes, but he's still wearing an expensive data center.

On Wednesday, the Federal Reserve cut rates by 25 basis points to 3.50%-3.75%, their third cut of 2025 and their last scheduled action of the year. Nine policymakers voted for it. Three voted against it—the most dissent since September 2019. That alone should have told you something. But equity markets did what they always do with a rate cut: they rallied on the headline and ignored the subtext. The Fed, in their statement, resurrected cautious language about "carefully assessing" future moves and signaled exactly one rate cut in 2026. Not two. Not three. One. Powell sat in his press conference and basically said the Fed is taking a vacation and will call you when they feel like coming back.

Then came Thursday, and the reckoning.

Oracle tanked 10% on Wednesday evening before Friday's open could even happen. Why? Because the company reported crushing debt-fueled AI capital spending—massive infrastructure buildout with delayed revenue recognition—while simultaneously missing guidance on top-line growth. Broadcom, on the other hand, posted genuinely impressive AI chip results: record $18 billion in quarterly revenue, up 28% year-over-year, with AI semiconductor revenue up 74%. Guidance was better than expected. The CEO practically promised the demand wave would continue into next year, projecting AI revenue to double to $8.2 billion in Q1.

So Broadcom crashed 11% on Friday anyway. Its worst day since January.

This is where the story gets interesting, because it reveals that the market isn't pricing in growth anymore—it's pricing in profitability, capital efficiency, and the dawning realization that "insatiable demand" was never the bottleneck. Execution was always the problem.

The AI narrative of 2024 and early 2025 was straightforward: build it and they will come. Hyperscalers were swimming in cash, rates were falling, returns-on-investment were theoretical but encouraging, and everyone from Nvidia to Oracle to Broadcom was seen as a vendor to the inevitable AI goldrush. The market accepted that thesis without much friction, because the alternative—admitting that much of the current valuation structure depends on speculative ROI assumptions—was too uncomfortable.

But something shifted. Maybe it was Oracle's warning that AI infrastructure capex is accelerating while revenue growth is decelerating. Maybe it was the cumulative effect of seeing the Magnificent Seven flatline while Russell 2000 index etfs hit new highs. Or maybe it's just that we're now far enough into the cycle that someone had to ask the obvious question: "When do these companies actually make money from this?"

The Fed's own actions may have inadvertently forced the conversation. Their "one cut in 2026" forecast, paired with higher growth and lower inflation expectations in their revised projections, signals something uncomfortable: they believe the economy is holding up fine without further relief. They cut this time because employment is weakening (initial jobless claims hit 236,000 last week, the highest in three months), but they're signaling they won't cut again unless something breaks. The "put" under equities that investors have grown accustomed to over the past five years just got a lot smaller.

Meanwhile, market expectations are already betting the Fed two cuts in 2026, according to the CME FedWatch tool—meaning traders aren't buying the Fed's own forecast. They're essentially saying "the labor market will deteriorate enough that you'll have to cut twice." That's either optimism about mean reversion or denial. Probably both.

Here's the piece that nobody wants to say out loud: Oracle's stock is down 40% from its September peak because investors finally did the math on the company's debt-to-capex trajectory. The company needed to show that massive spending on AI infrastructure would translate into immediate revenue growth and cash flow accretion. Instead, they showed that massive spending on AI infrastructure means more debt, delayed returns, and margin pressure in 2026. For a company like Oracle that doesn't have the balance sheet cushion of Alphabet or Microsoft, this math doesn't work at current valuations.

Broadcom's collapse despite excellent results points to an even darker possibility: that the custom-chip and semiconductor play—the "picks and shovels" bet that was supposed to be safer than owning the companies actually building AI—is also being repriced around something other than growth. Possibly around the reality that if demand truly is insatiable, then pricing power should be exceptional. If pricing power is normal, then demand probably isn't insatiable.

The sector rotation into small-caps and cyclicals we've seen this week isn't a sign of health. It's a sign of flight. High-valuation, high-growth narratives are losing their bid when the fed funds rate is staying put for months and the Fed has no intention of lowering it unless there's visible economic damage.

What happens next depends on whether the labor market crumbles fast enough to force the Fed's hand, or whether the economy proves sticky enough to keep rates where they are. If it's the former, AI stocks get a second wind but under darkening macro conditions. If it's the latter—if employment stabilizes around current levels—then we're looking at extended duration valuations in a period of dead money, possibly for years.

Powell said the Fed is "well positioned to wait and see." But waiting requires belief that the baseline holds. The market's sudden skittishness suggests belief is eroding faster than the Fed anticipated.

The AI reckoning wasn't supposed to happen like this—with rate cuts on the table and growth intact. But it's happening anyway. And the Fed's signal that cuts are largely over might have just hastened it.

Welcome to the phase where the rubber meets the road.