Two themes dominate today's semiconductor landscape: unprecedented AI data center infrastructure demand driving 56% revenue growth at MEP contractors, and a structural CPU demand surge tied to Intel's foundry turnaround under CEO Lip-Bu Tan. Semiconductor equipment names hit all-time highs while DeepSeek V4 benchmarks signal efficiency gains that could reshape inference economics. GPU allocation constraints persist at hyperscalers as AI model progress accelerates.
Comfort Systems USA reported Q1 2026 with 56.5% YoY revenue growth, ~51% same-store revenue growth, 88% Electrical segment growth, 47% Mechanical segment growth, and gross margin expansion to 26.3%. The call is a high-signal cross-market data point for AI infrastructure, data center electric load growth, and MEP construction capacity.
↳ Flex is positioned as the other key infrastructure play — supplying power shelves, busbars, rack-integrated cooling systems, and server builds directly to hyperscalers.
Intel CEO Lip-Bu Tan is credited for "continued and steady progress" in his first year as the AI data center boom drives demand. Lenovo's head of data center sales confirmed high-end server CPU supply is strained. Intel IR highlighted an unexpected margin lift: edge-die that would normally be scrap was downbinned and sold into usable SKUs — customers accepted all output.
↳ That level of demand acceptance is the defining characteristic of the current CPU cycle.
KLA Corporation (KLAC) is at an all-time high today with limited coverage. ASML is completely sold out and dependent on Carl Zeiss SMT capacity expansion. Both names benefit from broad AI infrastructure end markets.
↳ China overhang remains for KLAC but hasn't prevented the all-time high.
DeepSeek-V4 introduces 1.6T parameter Pro and 284B parameter Flash versions optimized for million-token context windows. Hybrid attention and Manifold-Constrained Hyper-Connections drive dramatic inference efficiency and memory management improvements.
↳ Open-weight frontier models with efficiency gains could pressure inference costs and reshape competitive dynamics.
OpenAI Chief Scientist Jakub Pachocki forecasts "significant improvements" in 1-2 months and "extremely significant improvements" in 3-6 months, calling recent progress "surprisingly slow."
↳ If Pachocki is correct, inference workload growth could accelerate, reinforcing GPU demand and semiconductor capex.
Santec (silicon photonics), FUSO Chemical (semi materials), Okano Valve (nuclear power), and Togami Electric (power distribution) are infrastructure beneficiaries. The Middle East conflict raises energy prices for European manufacturers, per Goldman Sachs Research.
↳ Nuclear power is increasingly viable for datacenter supply — a structural shift for power distribution names.
GPU useful lives are extending from 5 to 7-8 years with renewal prices rising. Nvidia holds ~75% gross margins while cloud clusters and memory suppliers see margin expansion. Smaller AI players face GPU allocation constraints — Microsoft Azure customers reportedly wait months for compute access.
↳ NVIDIA is oddly being left out of this CPU resurgence rally despite being the dominant GPU player.
Full analysis, implications, relative trade read-throughs, positioning reads and catalysts — delivered by email at 07:00 HKT daily.