AI Infrastructure —
Share
What changed in the AI trade, and why it mattered.
Networking and dc-infrastructure breadth collapsed while compute stayed strong, diverging within the AI trade.
Networking fell to 50% of names above the 200 EMA from 88% a week ago, and dc-infrastructure fell to 50% from 83%, while compute held at 100% (up from 89%) and servers/memory stayed at 100%. This is a rotation signal: the market is discriminating between AI infrastructure subsectors rather than trading the theme as one basket, with networking and power/cooling infrastructure names losing technical support even as chip and server names strengthen.
Compute · Semiconductor equipment · Memory · Networking · Servers · Data-centre infrastructure· QCOM, APH, GLW, CRDO, CIEN, COHR, VRT, TT, KLAC
Networking names split sharply: Amphenol crossed above its 200 EMA on a huge one-day move while Credo, Ciena, Coherent and Corning all broke down.
Amphenol's move stands out as an outlier even within a sector where four other names (CRDO, CIEN, COHR, GLW) crossed below their 200 EMAs in the last two days, reinforcing that networking breadth deterioration is broad-based rather than a single-name event, with Amphenol the exception.
Vertiv is acquiring Utility Innovation for up to $2.6 billion to expand AI data-center power capabilities, even as VRT crossed below its 200 EMA a day earlier.
The deal signals continued consolidation around power infrastructure for AI data centers, but the timing against a technical breakdown in the stock shows the acquisition news has not yet reversed the negative price trend for Vertiv.
Constellation Energy crossed above its 200 EMA and rose 3.5%, alongside continued reporting on Microsoft and Meta's multi-decade nuclear power deals.
Power remains a key bottleneck narrative for AI data centers; a nuclear-linked utility strengthening technically while hyperscalers lock in multi-decade nuclear supply agreements underscores where incremental capital is flowing within the power subsector, even though overall power breadth is flat at 50%.
Broadcom set a $230 billion AI revenue target through fiscal 2028, and TSMC's fab equipment demand has nearly doubled in six months with 2026 capex tracking toward $64 billion.
Both data points point to continued upstream capacity commitment in the semiconductor supply chain -- a foundry equipment shortage alongside a large multi-year revenue target from a major AI silicon supplier suggests demand visibility remains extended even as broader breadth narrows elsewhere.
Compute · Semiconductor equipment· AVGO, TSM· Source 1· Source 2· Source 3
Overall AI infrastructure breadth fell to 68% of names above the 200 EMA from 77% a week ago.
The broad-based decline, driven mainly by networking and dc-infrastructure weakness rather than compute or servers, indicates the AI trade is narrowing to fewer subsectors holding technical strength.
Theme-wide
Published 03/09/2026, 14:52 UTC · claude-sonnet-5