First Move  ·  AI Hardware & Infra  · 
The build-out — Wednesday morning, 22 July

AMD Secures Microsoft for Helios System, Challenges Nvidia at Advancing AI Summit

Investor focus shifts as AI chip valuations face scrutiny, while infrastructure bottlenecks and power demands drive new datacenter strategies.

ShareWhatsAppXLinkedIn

The story

This week, the competitive landscape in AI hardware intensified as AMD kicked off its 'Advancing AI 2026' summit in San Francisco. A significant development emerged just ahead of the event: AMD secured Microsoft as a new customer for its 'Helios' rackscale AI system.

This move directly challenges Nvidia's market dominance, as Helios has attracted major clients like Meta, OpenAI, and Oracle, demonstrating its total cost of ownership competitiveness despite a price tag roughly 40% higher than Nvidia's Vera Rubin. AMD is also expected to unveil detailed roadmaps for its next-generation 2nm server CPU, 'EPYC Venice,' at the summit.

This processor, based on the Zen 6 architecture, is slated to be the first high-performance computing chip manufactured using TSMC's cutting-edge 2-nanometer process and will feature up to 256 cores. The event underscores a broader industry shift where competition extends beyond individual chip performance to integrated system platforms and the entire AI infrastructure stack.

Silicon

MI350 series (MI350X, MI355X)

Maker: AMD

What: Features 288GB of HBM3e memory and delivers up to 4x performance on Llama 3.1 405B workloads compared to its predecessor.

For Whom: Microsoft Azure, Oracle, OpenAI, and Meta for AI inference and training workloads.

EPYC Venice

Maker: AMD

What: A 2nm server CPU based on Zen 6 architecture, featuring up to 256 cores, positioned as a high-performance computing processor.

For Whom: AI and HPC infrastructure deployments, unveiled at Advancing AI 2026.

NVLink 6 Switch

Maker: NVIDIA

What: The sixth generation of NVIDIA's purpose-built scale-up networking fabric, designed for high-bandwidth, low-latency GPU-to-GPU communication in AI factories.

For Whom: Datacenter operators to accelerate AI inference, training, and parallel computing workloads.

The build-out

Nebula Matrix AI LLC

Who: Senmiao Technology (AIHS) & Constant Energy Construction Corp. (CECC)

Scale: Joint venture to develop, invest in, construct, and operate AI data centers.

Where: United States

Beacon Point Phase 2

Who: Hut 8

Scale: 15-year, $9.8 billion AI data-center lease for compute capacity.

Where: Texas

AI-cloud contracts

Who: IREN

Scale: $2.8 billion in new multiyear AI-cloud contracts, raising year-end run-rate revenue target to over $4 billion.

Where: Global

Supply & policy signals

Investor rotation from AI chipmakers to hyperscalers

Implication: Hyperscaler capital expenditure growth is projected to slow sharply after 2026, from 76% this year to 25% in 2027 and 6% in 2028.

New York state imposes one-year moratorium on large new data centers

Implication: Mounting concerns over environmental and infrastructure costs, including electricity consumption and water usage, are leading to regulatory action.

China considering new AI export controls

Implication: Potential restrictions could prevent overseas fabs like TSMC from producing advanced chips designed by Chinese firms, further bifurcating the global AI chip market.

TSMC increases 2026 capital spending plans

Implication: Strong, multi-year demand for AI chips is driving continued investment in advanced manufacturing, despite geopolitical tensions.

What we'll be watching

Reporting + analyst voices: grounded via Google Search at publish time.