Billionaire Ken Griffin Is Shifting His AI Bets. Here's What He's Buying and Selling.

Citadel reduced AI-chip exposure by selling 6.8M Nvidia shares (cutting exposure by 36%), trimmed Broadcom (~11%) and Micron (~87%), while adding to Amazon and Microsoft. Amazon posted ~$200B quarterly revenue (+~20% YoY) with AWS operating income rising to $16.6B from $10.2B and AWS recording its fastest growth in 4+ years; Microsoft delivered $19.6B free cash flow in Q4 FY2026 and expects to stay free-cash-flow positive in FY2027. Since reporting late July earnings, Amazon and Microsoft are up ~15% and ~27%, respectively, supporting a broadly positive read-through on results from AI spend.

Analysis

The market implication is not “AI is over”; it is that capital is migrating from the picks-and-shovels leg of the cycle to the monetizers. Hyperscalers with visible cash generation can keep funding AI capex without the balance-sheet scar tissue that usually accompanies a platform land grab, so the next rerating should favor names that can show both usage growth and free-cash-flow durability. That creates a second-order squeeze on component suppliers: if the spend pool gets more disciplined, semis become more exposed to quarterly digestion and multiple compression even if end-demand stays healthy.

AMZN and MSFT are the cleaner beneficiaries because their AI spend can be justified by customer lock-in, attach rates, and operating leverage rather than pure unit shipment growth. The subtle read-through is negative for NVDA/AVGO/MU not because their fundamentals are broken, but because investor willingness to pay peak-cycle multiples for infrastructure growth is fading as the “show-me” phase matures. Over the next 1-3 months, the key catalyst is whether cloud reacceleration persists into guidance and whether capex converts into margin expansion; if not, the market will start discounting ROI slippage instead of AI optionality.

Contrarian view: the selloff in chip exposure may be overinterpreted as a bearish call when it may simply reflect crowding and factor management. If hyperscaler capex keeps rising faster than depreciation, the semis can still outperform on earnings revisions even while long-only positioning rotates away from them. The real risk is 6-18 months out: if AI workloads fail to improve enterprise monetization, the current spend curve becomes a margin tax on AWS/Azure and the winners revert to the hardware layer with the best pricing power. The thesis is falsified if AMZN or MSFT guide cloud growth deceleration or if incremental AI investment stops lifting operating income and free cash flow.

Key Decisions for Investors

• Long AMZN / long MSFT on any post-earnings pullback over the next 1-3 weeks; favor entries only if both hold above their post-report breakout levels, because the trade is a proof-of-monetization story rather than a valuation trade.
• Pair trade: long AMZN + MSFT vs short NVDA + AVGO + MU or a semicap basket (SMH) into the next cloud/AI earnings window; target is relative multiple compression in hardware if capex growth remains strong but selective.
• If already long semis, trim 20-30% into strength and rotate into hyperscalers; the best risk/reward now is not chasing the most levered capex names before confirmation of spending durability.
• Watch the next Azure/AWS commentary and capex guidance as the falsifier: if cloud revenue growth re-accelerates while FCF stays positive, add to AMZN/MSFT; if margins slip or capex guidance is cut, cover longs and re-add NVDA on weakness.