Will any hyperscaler announce in-house power or cooling manufacturing capability by end of 2026?

The question asks about in-house manufacturing capability, not design. This is a critical distinction. Hyperscalers may design custom cooling or power solutions (Amazon already has), but building dedicated manufacturing facilities for power conversion or liquid cooling is a fundamentally different capability. Manufacturing at scale requires massive capital investment, supply chain management, and years of operational expertise that hyperscalers do not possess. Their core competence is software and services, not hardware manufacturing. In 2026, they are more likely to deepen partnerships with Flex than build competing facilities.

Even Amazon, the most vertically integrated hyperscaler (custom chips, custom networking), has not moved into power or cooling manufacturing. The Amazon cooling announcement was about design specifications, not manufacturing. Building power conversion equipment (800V DC converters, switchgear) requires specialized engineering talent and regulatory certifications. Liquid cooling manufacturing (cold plates, CDUs) requires precision manufacturing. Neither aligns with hyperscaler core competencies. The probability is low.

I assign slightly higher probability because the question spans through end of 2026 and includes four hyperscalers. The probability isn't just about current plans -- it includes the possibility that one hyperscaler acquires a power or cooling manufacturer (rather than building in-house). An acquisition of a liquid cooling startup or power equipment company would technically count as 'announcing in-house capability.' Amazon, Google, or Microsoft have the capital to make such acquisitions. This increases the probability modestly.

Amazon's cooling announcement was a signal, not an endpoint. Hyperscalers have been gradually internalizing more of the data center stack. Custom chips (Google TPU, Amazon Graviton), custom networking, now custom cooling. Power may be next on the roadmap. The question is whether any announcement happens by end of 2026 -- which is only 9 months away. I think this is more of a 2027-2028 event, but I assign meaningful probability because the trend is clearly toward internalization.

Manufacturing is fundamentally different from design. Hyperscalers have shown willingness to design custom components but consistently partner for manufacturing (TSMC for chips, Flex for servers, etc.). The economics of manufacturing don't align with hyperscaler business models -- they prefer asset-light approaches with contracted manufacturing partners. An announcement of in-house manufacturing would be unprecedented and unlikely in the 2026 timeframe.

The resolution criteria specifies 'plan to build in-house manufacturing capability' -- not actual operational manufacturing. An announcement of plans is a lower bar than actual construction. A hyperscaler could announce a pilot manufacturing facility or an acquisition of a cooling/power company, which would count. This slightly increases the probability compared to actual manufacturing capability. Still unlikely in 2026 but not negligible.

Hyperscalers design, partners manufacture. This model has held for decades. Unlikely to change in 9 months. Low probability.

Building manufacturing takes years. Even if a hyperscaler decided today to build in-house power manufacturing, they wouldn't announce it in 2026 as a capability -- maybe as a future plan. The 'plan to build' language in resolution criteria is somewhat more lenient but still very unlikely.

The acquisition angle increases the probability. If Amazon or Google acquires a liquid cooling or power equipment company, it could count as in-house capability. Tech companies frequently make strategic acquisitions. This is the most plausible path to YES. Still minority probability.