Infrastructure evolution over time — what's changing across fabs, chips, nodes, and datacenters.
This Week
1
infrastructure changes
Significant
9
high-impact events
Active Alerts
5
live infrastructure alerts
30-Day Changes
8
tracked events
INFRASTRUCTURE CHANGES — CHRONOLOGICAL
TSMC revised CoWoS-L capacity target upward to ~50K wspm for 2025, reflecting accelerating AI demand from NVIDIA and AMD.
TSMC Arizona Fab 21 Phase 1 (N4P) entered volume production — milestone for US advanced semiconductor manufacturing.
NVIDIA Blackwell B200 GPU reached general availability — CoWoS-S packaging at scale, HBM3e 192GB per GPU.
Intel 18A process node achieved power-on for first external customer test chip — key milestone for Intel Foundry viability.
Northern Virginia power grid utilization rose to 87% as hyperscale AI cluster buildout accelerated — grid stress increasing.
Google announced expanded TPU v5e deployments across GCP regions — AI inference workload shift accelerating.
TSMC N2 mass production confirmed for H2 2025 — initial customers include Apple (A19), NVIDIA (next-gen). On-track with prior guidance.
SK Hynix confirmed HBM4 sample delivery to NVIDIA ahead of schedule — 1.2 TB/s bandwidth, 6-hi/8-hi stacks planned.
Intel Ohio fab construction continues on revised schedule — Phase 1 now targeting 2026 completion vs. original 2025 plan.
GlobalFoundries expanded partnership with US DoD for secure, radiation-hardened chip manufacturing at Malta NY fab.
Microsoft Redmond AI campus power capacity upgraded to 1.2 GW as Blackwell GPU cluster deployment accelerated.
Intel 18A (1.8nm-class) ribbon FET gate-all-around transistors confirmed in production test — ahead of Samsung's GAA timeline.
ACTIVE INFRASTRUCTURE ALERTS
CoWoS Capacity Pressure Elevated
Advanced packaging capacity pressure for AI GPU CoWoS orders remains elevated heading into H2 2025. TSMC CoWoS-L allocation tightly managed.
Entity: CoWoS-L
HBM Supply Tightening for 2025-2026
HBM3e and early HBM4 supply constrained by SK Hynix and Micron capacity limits. AI training cluster buildouts driving demand beyond current capacity.
Entity: HBM Integration
Taiwan Concentration Risk Remains Critical
Over 90% of leading-edge AI chip production concentrated at TSMC Taiwan fabs. No credible alternative at N3/N4 scale exists before 2027.
Entity: TSMC
AI Datacenter Power Demand Surpassing Grid Projections
Hyperscale AI cluster power demand in Northern Virginia and Silicon Valley exceeding utility projections by 18-24 months — grid expansion lagging buildout.
Entity: Northern Virginia
Intel Ohio Fab Phase 1 Delayed to 2026
Intel's Ohio fab Phase 1 completion pushed from 2025 to 2026. Reduces near-term US-based leading-edge capacity diversification.
Entity: Intel Ohio Fab
SIGNIFICANT EVENTS
TSMC revised CoWoS-L capacity target upward to ~50K wspm for 2025, reflecting accelerating AI demand from NVIDIA and AMD.
5 days ago
TSMC Arizona Fab 21 Phase 1 (N4P) entered volume production — milestone for US advanced semiconductor manufacturing.
1w ago
NVIDIA Blackwell B200 GPU reached general availability — CoWoS-S packaging at scale, HBM3e 192GB per GPU.
1w ago
Intel 18A process node achieved power-on for first external customer test chip — key milestone for Intel Foundry viability.
2w ago
Northern Virginia power grid utilization rose to 87% as hyperscale AI cluster buildout accelerated — grid stress increasing.
2w ago
TSMC N2 mass production confirmed for H2 2025 — initial customers include Apple (A19), NVIDIA (next-gen). On-track with prior guidance.
3w ago
Demo data. Change events are illustrative examples of what this tracker surfaces. Live ingestion populates this feed automatically as sources are reviewed and published.