AI Infrastructure —
Share
What changed in the AI trade, and why it mattered.
Compute breadth hit 100% above the 200DMA, up from 89% a week ago, led by Qualcomm's cross.
A full sweep in compute alongside Qualcomm's move above its 200DMA (aided by the Nvidia-MediaTek tie-up narrative) signals broad-based technical strength in the segment just as AVGO reports next week; power is moving the opposite way, showing the divergence is now segment-specific rather than market-wide.
Nvidia is investing $3.5 billion in MediaTek and partnering on an NVLink Fusion platform.
This extends Nvidia's strategy of embedding its interconnect standard into Big Tech's custom silicon buildout, a read-through for how much of the AI chip supply chain remains tied to Nvidia's ecosystem even as hyperscalers pursue in-house silicon.
Power breadth fell to 42% above the 200DMA from 50%, with Siemens Energy and Schneider Electric both down sharply and Siemens Energy crossing below its 200DMA; dc-infrastructure names split, with Vertiv crossing above and Trane crossing below.
Power is now the weakest segment in the coverage universe on breadth, and the split within dc-infrastructure suggests the power buildout trade is becoming more idiosyncratic than a uniform tailwind, worth watching given Anthropic's reported 5GW capacity secure and AMD's Saudi gigawatt-scale build.
Anthropic agreed to a $35 billion cloud contract with Lambda.
A deal of this size for a neocloud provider underscores how much frontier-model compute demand is flowing outside the traditional hyperscaler channel, a direct read for neocloud capacity economics and GPU demand pull-through.
Neoclouds · Hyperscalers · Compute· Source
Microsoft's contracted backlog stands at $678 billion and Oracle's at $638 billion against a $430 billion market cap, while Goldman Sachs is flagging an upcoming AI capex surprise for AMZN, GOOGL, META, and MSFT.
The scale of contracted backlog relative to market cap (especially at Oracle) plus a fresh capex surprise call from Goldman keeps the hyperscaler spending trajectory as the dominant demand signal for the whole AI infrastructure chain, even as one story flags this backlog as a source of systemic-risk concern.
Hyperscalers · Compute · Servers· MSFT, ORCL, AMZN, GOOGL, META· Source 1· Source 2· Source 3
China's CXMT reportedly made a major AI memory chip breakthrough, a headline risk for Micron and Sandisk even as Sandisk closed 5.5% higher on the day.
A credible domestic Chinese competitor in AI memory would pressure the pricing power and supply discipline that has underpinned the memory rebound trade; the divergence between the bearish headline and Sandisk's daily move is worth flagging rather than reconciling.
Unimicron, a key PCB supplier to Nvidia and Intel, is under investigation for alleged China-origin fraud with potential 40% U.S. tariff exposure.
A tariff penalty at that scale on a shared supplier would ripple into component costs and supply timelines for both Nvidia and Intel, a tail risk for the physical supply chain underpinning compute buildouts.
Overall AI infrastructure breadth held flat at 77% above the 200DMA, with semicap, memory, networking, and servers all unchanged at 100%.
Most of the coverage universe showed no technical change this week; the real movement is concentrated in compute's breadth jump and power's breadth decline, making this a quiet week outside those two poles.
Semiconductor equipment · Memory · Networking · Servers · Data-centre infrastructure · Hyperscalers · Neoclouds
Published 01/09/2026, 09:20 UTC · claude-sonnet-5