Ken Griffin's Citadel Rotates Out of AI Infrastructure Into AI Profits

Ken Griffin has built Citadel into one of the largest multi-strategy funds on the planet by moving early and moving decisively. His firm's Q2 2026 13F filing, covering positions as of June 30, shows exactly that kind of decisive move: a rotation away from the chipmakers that sell AI infrastructure and toward the companies that are actually turning AI spending into cash flow.

What Citadel sold

The filing shows Griffin trimming three of the AI trade's most crowded names:

Nvidia took the biggest cut in dollar terms — Citadel sold roughly 6.8 million shares, a 36% reduction, though the position remains large enough to stay among the fund's top five holdings. Broadcom was pared by close to 11%, or about 356,000 shares. The most dramatic exit was in Micron Technology, where Citadel dumped 4 million shares — 87% of its prior stake — effectively walking away from the memory-chip trade almost entirely.

None of this means Griffin has turned bearish on AI. It reads more like profit-taking after a run in which all three stocks benefited enormously from the buildout narrative. Investors tracking crowded, multi-fund positions like these can see how exposed the broader hedge fund universe still is to the same handful of chip names on InvestorLens's portfolio overlap tool, which flags when several top managers are piled into the same trade at once — exactly the kind of setup that tends to unwind quickly once one large holder starts selling.

What Citadel bought instead

The proceeds went somewhere specific: companies with AI businesses that are already generating measurable revenue and cash flow, rather than companies selling the picks and shovels.

Amazon was the standout addition. Citadel bought 2.8 million more shares, lifting its stake to 10.9 million shares worth about $2.6 billion, or roughly 1.49% of the fund's reported portfolio. Microsoft got a similar treatment — Griffin added 1.1 million shares, bringing the total to 3.8 million shares valued near $1.4 billion, about 0.81% of the portfolio.

The rationale lines up with what both companies reported in their most recent earnings. Amazon's AI-related business has reportedly reached a $25 billion annualized run rate, giving Citadel a concrete revenue number to point to rather than a narrative about future demand. Microsoft, meanwhile, generated $19.6 billion in free cash flow in the period even while running roughly $175 billion in annualized AI infrastructure spending — proof, in Griffin's apparent view, that the company can fund its own buildout without straining the balance sheet. Since both companies reported earnings in late July, Amazon shares are up about 15% and Microsoft is up around 27%, which means Citadel's Q2 addition was already working well before the filing became public in mid-August.

The bigger pattern: infrastructure spenders vs. infrastructure earners

This filing fits a theme that's shown up elsewhere in 13F data this year: institutional money increasingly separating "AI capex" stocks from "AI monetization" stocks. Chipmakers and memory suppliers sell the shovels, but their upside is tied to the capital-spending cycle continuing at an extraordinary pace — a pace several fund managers have started to question. Hyperscalers like Amazon and Microsoft, by contrast, get to show both sides of the ledger: what they're spending, and what it's already generating back.

It's worth noting Citadel is a multi-strategy market maker as much as it is a directional stock-picker, so its 13F doesn't map cleanly onto conviction the way a concentrated fund like Pershing Square's might — some of this exposure is likely hedged or offset elsewhere in the firm's book. Still, a rotation this size, in this direction, is a useful data point for anyone trying to read where institutional sentiment on AI is heading next. For a broader read on how positioning is shifting across the hedge fund universe heading into year-end, InvestorLens's fund flow tracker and macro consensus dashboard both pull in the latest 13F-based signals as new filings land.

Anyone who wants to see exactly how Griffin's portfolio compares to other prominent AI investors — or track his position changes filing over filing — can pull up Citadel's full holdings history on the investor profiles page.

The takeaway

Griffin didn't sell AI in Q2 2026 — he sold the part of AI that's still mostly a promise and bought more of the part that's started cashing checks. Whether that rotation ages well depends on whether Nvidia, Broadcom, and Micron can show their own monetization story in upcoming quarters, or whether the chip names keep correcting while hyperscaler earnings keep climbing. Either way, it's a reminder that 13F filings are most useful not as buy signals on their own, but as a record of how the smartest allocators in the market are adjusting their risk as the AI trade matures.

---

Data sourced from public SEC 13F filings. Educational research only — not investment advice.