Major Hyperscalers Just Reported Earnings. Nvidia Was The Winner

Alphabet, Microsoft, and Amazon signaled continued heavy AI capex, with Amazon projecting $200 billion in capex this year and Microsoft guiding to $190 billion in 2026. The article argues this spending supports Nvidia demand, especially as the company’s two largest direct customers accounted for 36% of revenue in the latest fiscal year. Nvidia remains viewed as attractive due to its market-leading hardware, CUDA switching costs, and upcoming Vera Rubin platform.

Analysis

The market is still underestimating how sticky AI infra spend becomes once the first wave of model deployment moves from experimentation to service-level commitments. The more important signal is not that cloud hyperscalers are spending, but that they are each rationally spending in parallel, which reduces the odds of a near-term capex pause and keeps pricing power with the best-performing training silicon. That dynamic favors the incumbent GPU leader because the buying decision is increasingly about time-to-capacity and ecosystem lock-in, not just unit cost.

Second-order, the biggest pressure lands on custom silicon narratives. Even if internal accelerators gain share at the margin, they mainly cap Nvidia's upside in the low-end workload bucket rather than dislodge it from the frontier-training segment where performance per watt, software maturity, and developer familiarity matter most. The supply-chain beneficiaries extend beyond NVDA to advanced packaging, HBM, and optical interconnect names; those are the more underappreciated bottlenecks if hyperscaler capex keeps inflecting into 2026.

The main risk is timing, not thesis. If enterprise AI monetization disappoints over the next 2-3 quarters, cloud vendors could slow the pace of incremental deployment and stretch utilization on existing clusters, which would hit order growth before it hits revenue. A secondary risk is a transition from scarcity to procurement discipline once next-generation in-house chips become available, but that is a 12-24 month issue, not a near-term catalyst against the group.

Consensus is too focused on whether Nvidia loses share to custom chips and not enough on how a multi-year capex arms race expands the total market. If the spend continues, the bigger trade is not just NVDA upside, but a widening performance gap between the AI compute supply chain and the rest of megacap tech. In that setup, pullbacks should be bought on any evidence that cloud capex remains above depreciation by a wide margin.

Key Decisions for Investors

• Stay long NVDA on weakness over the next 2-6 weeks; use any post-rally consolidation to build exposure because the next 2 quarters likely still show order visibility exceeding street expectations. Risk/reward remains favorable as long as hyperscaler capex commentary stays constructive.
• Pair trade: long NVDA / short INTC for 3-6 months. The thesis is that custom silicon can compete on cost but not on ecosystem depth or frontier-performance relevance; risk is a sharper-than-expected improvement in Intel’s AI accelerator execution.
• Add exposure to the AI picks-and-shovels basket into any market dip: long GOOGL, AMZN, MSFT plus select beneficiaries in advanced packaging/HBM supply chains. The trade works best if you expect capex to stay elevated through 2026 and cannot be cleanly hedged by shorting the hyperscalers.
• If you want convexity, buy NVDA 9-12 month call spreads rather than outright calls. This expresses continued AI spend with defined downside if procurement shifts to internal chips faster than expected.
• Reduce or hedge NVDA only if a major cloud provider explicitly guides to slower 2026 data-center capex for two consecutive quarters; absent that, the bear case is premature.