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 about $725 billion on capital expenditures in 2026, up roughly 77% year over year, underscoring continued AI infrastructure demand. The article argues this is a mixed setup for Nvidia: hyperscalers are building custom chips to reduce dependence over time, but they are still buying record volumes of Nvidia GPUs and Nvidia data center revenue rose 92% year over year to $81.6 billion in its latest quarter. The near-term impact is more about valuation and share mix risk than an immediate demand collapse.

Analysis

The key market implication is not that hyperscalers stop buying Nvidia; it is that the mix of spend gradually shifts from merchant silicon to captive silicon while total AI infrastructure budgets keep expanding. That creates a slower, more painful version of share loss for NVDA: not a demand cliff, but a margin-compression story as price discipline weakens once customers can credibly redirect a portion of workloads to in-house chips. The market is likely still underestimating how much of the next leg of AI capex will be “good enough” silicon plus proprietary software, rather than best-in-class GPUs for every workload.

The second-order winner is the cloud platform layer and the adjacent networking / interconnect stack, because custom chips do not eliminate the need for racks, power, cooling, optics, and high-speed networking. In other words, even if some compute dollars migrate away from NVDA, the beneficiaries are not automatically the hyperscalers alone; the infrastructure bill still has to be paid to data-center landlords, electrical gear vendors, and high-bandwidth networking suppliers. That means the real competitive threat to NVDA is more likely to show up in gross margin and mix than in top-line collapse over the next 12-24 months.

The contrarian miss is timing: investors may be pricing a structural Nvidia share loss as if it is imminent, while the transition is actually gated by software maturity, reliability, and deployment cycles. Custom silicon typically takes multiple generations before it can displace merchant GPUs across training and inference, so the bear case can be right directionally but early in the earnings timeline. The more interesting risk for NVDA is that the market keeps assigning it a premium multiple for scarcity just as the scarcity premium begins to erode.

For AMZN, GOOGL, and MSFT, the strategic value of in-house chips is less about replacing Nvidia overnight and more about improving bargaining power, workload allocation, and long-term cloud margins. If any one of these companies proves it can expose its own chips externally at scale, it creates a new profit pool and a new pricing benchmark for AI compute. That would be a bigger threat to NVDA than internal substitution alone, because it turns custom silicon from a cost-saving tool into a platform business.

Key Decisions for Investors

• Reduce NVDA exposure on strength over the next 1-3 months; the risk/reward is now skewed toward multiple compression rather than earnings collapse. Use rallies to trim, not sell into weakness, because the near-term numbers can still print well.
• Pair trade: long GOOGL / short NVDA for 3-6 months. Thesis: Alphabet has optionality from TPU commercialization while NVDA faces the first meaningful narrative pressure on pricing power; target asymmetric downside in NVDA if hyperscaler commentary turns incremental.
• Pair trade: long AMZN / short NVDA into the next capex cycle. Amazon has the cleanest path to monetizing custom silicon while still absorbing huge AI spend; the spread should benefit if investors start rewarding capex efficiency rather than pure compute scale.
• Consider a basket long of AI infrastructure enablers versus NVDA, focused on power, cooling, and networking names not named here, for a 6-12 month window. If compute mix shifts, the capex dollar does not disappear—it gets repriced across the supply chain.
• Use NVDA puts or put spreads on post-earnings/ post-guidance rallies as a tactical hedge for 2-4 months. The setup favors slow decay in market share expectations rather than a one-quarter miss, so defined-risk downside exposure is preferable to outright shorting.