AI Infrastructure —
Share
What changed in the AI trade, and why it mattered.
UBS raises its AI capex forecast to nearly $1 trillion this year and about $1.4 trillion by 2027
A bigger third-party capex forecast reinforces the demand backdrop for every layer of the AI infrastructure trade, from chips and optical gear to power and data centers, and helps explain why individual capacity announcements are landing well.
Theme-wide· Source
Equinix commits $5 billion to $7 billion a year to data center buildout, and Oracle's hundreds of billions in AI data center commitments are flowing through to industrial suppliers
Sustained, large-scale capex from a colocation leader and a hyperscaler-adjacent cloud player signals that AI-driven data center demand is being funded for years out, benefiting builders, equipment makers, and power and cooling suppliers down the chain.
Data-centre infrastructure · Hyperscalers· EQIX, ORCL· Source 1· Source 2
Ciena targets $14 billion in revenue by 2029 as it says optical networking demand is outrunning supply
A supply-constrained demand backdrop in optical networking points to pricing power and multi-year visibility for the networking layer that connects AI data centers, a read-through for other optical and interconnect suppliers.
GlobalFoundries jumps 7% on an expanded Marvell agreement
An expanded foundry agreement tied to Marvell signals deepening custom silicon demand tied to AI infrastructure, a positive read for semicap and foundry capacity utilization.
CoreWeave is raising prices on AI compute and customers are still paying
Pricing power in AI compute rental markets suggests demand is outstripping available capacity, a bullish signal for neocloud economics and for GPU suppliers whose hardware underpins that compute.
GE Vernova's $200 billion order backlog is being fulfilled faster than expected, and Nvidia has formed an alliance to build a power-flexible AI data center
Faster backlog conversion at a major power equipment supplier alongside Nvidia's push into power-flexible data center design underscores that electricity supply, not just chips, is becoming a critical bottleneck and investment focus in AI infrastructure.
China's CXMT moves its new memory-chip platform into mass production
A domestic Chinese memory supplier reaching mass production adds a new source of memory capacity outside the established suppliers, a development worth watching for global memory pricing and supply dynamics even though memory names have been uniformly strong.
Theme-wide· Source
Semicap and power names broadened out this week while networking also strengthened, even as compute, memory, servers, hyperscalers and neoclouds held steady
The move higher in semicap, power and networking was not confined to one or two names, with several stocks in each group turning strong for the first time in a week, suggesting the AI infrastructure advance is widening beyond the compute and memory names that had already been firm.
Compute · Semiconductor equipment · Memory · Networking · Servers · Power · Hyperscalers · Neoclouds· AMAT, KLAC, ONTO, GLW, COHR, GEV
Published 20/09/2026, 16:26 UTC · claude-sonnet-5