Semiconductors peak as hyperscalers attempt to catch up

For nearly a year, semiconductor stocks have been eating hyperscaler stocks for breakfast. Since September 2025, the companies making AI chips have dramatically outperformed the companies buying them. Now, according to JPMorgan analysts, that trade may be running out of runway.

The core tension is elegant in its simplicity. Hyperscalers like Microsoft, Amazon, Google, and Meta have been writing enormous checks to semiconductor suppliers for AI infrastructure. Those suppliers have been happily cashing them. But the “check-writers” are getting tired of being on the wrong side of the equation, and they’re doing something about it.

The $800 billion question

Hyperscalers are projected to spend roughly $800 billion on AI infrastructure in 2026. Some forecasts put 2027 spending as high as $1.1 trillion. Those are staggering numbers, and they’ve been a massive tailwind for semiconductor companies that supply the chips, memory, and networking gear powering the AI buildout.

But here’s the thing. That same capital is increasingly flowing toward a strategy designed to cut semiconductor suppliers out of the loop entirely.

Google is pushing forward with its TPU v6 chips. Amazon has Trainium2 and Inferentia3 in its pipeline. The industry shorthand for this shift, “check-writers vs. check-receivers,” started circulating in early July 2026 commentary. It captures a simple but powerful idea: the companies spending the most money on AI are beginning to build leverage over the companies taking that money.

Why semis outperformed in the first place

The semiconductor outperformance since September 2025 wasn’t irrational. High-bandwidth memory pricing surged sharply during this period, creating significant pressure points across the semiconductor supply chain. When your product is in short supply and your customers are in a spending arms race against each other, you have pricing power.

Analysts at both JPMorgan and Nomura have pointed to ongoing hyperscaler capital expenditure commitments and supply chain bottlenecks as reasons semiconductor stocks held up even as peak-performance concerns grew.

The vertical integration threat is real this time

Google’s custom TPU chips are already handling significant inference workloads. Amazon’s Trainium2 is designed for training large models at lower cost than third-party GPUs. Meta has been developing its own inference accelerators. These aren’t PowerPoint slides. They’re shipping products with real workloads running on them.

The multi-vendor GPU strategy that several hyperscalers have adopted is another quiet threat. By qualifying chips from AMD, Intel, and smaller players alongside Nvidia, hyperscalers create competitive tension among suppliers.

What this means for investors

Short-term, semiconductor stocks may continue to benefit from supply chain tightness and memory pricing dynamics that haven’t fully normalized. The HBM pricing surge alone has been a significant earnings driver, and those contracts don’t unwind overnight.

The anticipated hyperscaler expenditures exceeding $800 billion in 2026 will continue to bolster demand for certain semiconductor firms, particularly those with differentiated technology or deep integration into hyperscaler supply chains. JPMorgan’s characterization of the semiconductor outperformance as “potentially unsustainable” is diplomatic language for what could be a meaningful rerating.

Tracking the progress of Google’s TPU v6, Amazon’s Trainium2 and Inferentia3, and similar custom silicon programs is the single most important variable for anyone holding semiconductor positions into 2027.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.