AI Infrastructure —
Share
What changed in the AI trade, and why it mattered.
AWS tells clients to abandon Middle East data centers after Iranian strikes left facilities inaccessible for six months
A hyperscaler retreating from a region with no recovery timeline is a direct hit to cloud capacity buildout plans and raises questions about geographic concentration risk in AI infrastructure, with ripple effects for vendors like Cisco and Oracle tied to those sites.
Networking · Hyperscalers· AMZN, CSCO, ORCL· Source 1· Source 2
Vertiv landed a $2.6 billion AI power deal while GE Vernova, Eaton and Quanta Services all rose on backlog and demand signals
Power and cooling infrastructure remains a bottleneck for AI data center buildout, and these deals show continued capital commitment even as the power sector overall has been weakening.
Data-centre infrastructure · Power· VRT, GEV, ETN, PWR· Source 1· Source 2
Broadcom CEO Hock Tan guided AI revenue to $230 billion by 2028
This is one of the most aggressive multi-year AI revenue targets from a major semiconductor supplier and signals confidence in custom silicon and networking demand from hyperscalers.
SK Hynix reportedly in talks with Intel to build memory chips in the US, and Coherent, Intel and Marvell all posted strong single-day gains
A potential SK Hynix-Intel manufacturing tie-up would reshape US memory supply chains and reduce reliance on Asian fabs, while the price moves show renewed strength concentrated in a few compute and networking names.
Compute · Memory · Networking· INTC, 000660.KS, COHR, MRVL· Source
CoreWeave signed a 15-year lease with Blockfusion at a Niagara Falls AI campus, and Shell struck a $715 million power supply deal with Constellation Energy
Long-duration leases and power supply agreements underscore continued long-term capacity commitments in the AI buildout even as broader sector strength has been softening.
Huawei set a 2027 launch target for new AI chips aimed at competing with Nvidia
A concrete timeline from Huawei signals China's AI chip self-sufficiency push is advancing, a slow-building competitive risk to Nvidia's dominance in AI accelerators, particularly in China-adjacent markets.
Semicap and power names weakened sharply this week while compute, memory and servers held firm
The pullback in semicap, where nearly all names fell below their trend after being universally strong a week ago, and in power, contrasts with steady strength in compute, memory and servers, suggesting the AI trade's weakness is concentrated rather than broad-based.
Compute · Semiconductor equipment · Memory · Servers · Power· AMAT, ONTO, KLAC, GEV
Published 17/09/2026, 07:01 UTC · claude-sonnet-5