Amazon, Alphabet, and Microsoft Are All Racing to Design Their Own AI Chips. Here's What It Means for Nvidia.

Amazon, Alphabet, Microsoft, and Meta are on track to spend roughly $725 billion on capex in 2026, up about 77% from last year, while building more custom AI silicon to reduce reliance on Nvidia. Even so, Nvidia’s latest fiscal first-quarter revenue rose 85% year over year to $81.6 billion, with data center revenue up 92% and hyperscalers still contributing about half of that business. The article frames the outlook as mixed: custom chips could pressure Nvidia’s pricing power over time, but overall AI spending is still expanding fast enough to support continued growth.

Analysis

The market is still treating AI capex as a single trade, but the composition is fracturing: hyperscalers are increasingly buying from themselves first and Nvidia second. That creates a subtle margin transfer from the chip layer to the cloud/platform layer over the next 12-36 months, but it does not imply a clean revenue cliff for NVDA because the total installed base of AI workloads is still expanding faster than custom silicon can displace it. The bigger risk for Nvidia is not volume collapse; it is that mix shift gradually compresses pricing power while investors continue to capitalize the business as if share and margin are both permanent.

The second-order winner is the vertically integrated cloud stack: AMZN, GOOGL, and MSFT can convert capex into lower unit economics, better internal utilization, and eventually external chip/services revenue. Alphabet has the cleanest optionality because TPUs can become a cloud product with a scarcity premium when compute is rationed; Amazon has the deepest self-help runway because custom silicon directly lowers AWS infrastructure cost. Microsoft is the least advanced, so its near-term capex is still the most Nvidia-linked, which makes it the most exposed if capacity constraints persist while AI demand stays elastic.

Contrarianly, the consensus underestimates how long Nvidia can grow even as it loses share at the margin. The stock’s risk is less “custom chips win” and more “custom chips win slowly,” which would allow earnings to keep rising but at a decelerating multiple expansion rate. That is a setup for mean reversion in NVDA’s valuation premium rather than an immediate fundamental break. The tail risk for the bears is that non-hyperscaler demand from startups, enterprises, and governments becomes the dominant incremental buyer, offsetting any hyperscaler substitution for several quarters.

Key Decisions for Investors

• Short NVDA vs long a basket of AMZN/GOOGL/MSFT on a 3-6 month horizon; thesis is relative margin transfer and slower multiple support for NVDA, with stop-loss if Nvidia re-accelerates data center growth above current trend.
• Add/overweight AMZN and GOOGL into any AI-semiconductor selloff over the next 1-2 weeks; both have the clearest path to lowering effective compute cost per unit of AI output while retaining upside from externalizing custom silicon.
• Buy MSFT on weakness only as a lagging beneficiary, but hedge with NVDA puts or a semiconductor ETF overlay for 1-2 quarters; MSFT remains the most dependent on third-party accelerators near term, so the best entry is on disappointment rather than momentum.
• Use NVDA call spreads, not outright longs, for the next earnings cycle; the setup favors continued growth but rising debate on peak share, so capped upside reduces valuation risk if multiple compression starts before fundamentals roll over.
• Watch for any public evidence of TPU/Trainium/Maia externalization becoming revenue-bearing in the next 6-12 months; that is the real catalyst that would shift this from a cost-saving story into a direct competitive assault on Nvidia's gross margin structure.