- About Us
-
Research Report
Semiconductors
LED
Consumer Electronics
Emerging Technologies
- Selected Topics
- Membership
- Price Trends
- Press Center
- News
- Events
- AI Agent
NVIDIA advances its 800V Power Rack roadmap, with mass adoption expected from Rubin Ultra onward, while North American data center buildouts face delays from grid and equipment constraints.
As AI shifts from training to inference and agentic workloads, Server CPU evolves from auxiliary to core orchestrator, lifting both ARM and x86 demand into a new growth cycle.
The year 2026 marks the watershed transition of AI data center optical interconnects from adoption to mass production. The actual volume driver is not scale-out CPO switches, but GPU scale-up networks. Competition is also elevating from pure PIC foundries to integrated optical engine platforms that combine PICs, EICs, packaging, lasers, and optical coupling. Success in mass production will depend on yield, field-serviceable connectors, InP/CW laser supply, and strategic ecosystem positioning.
AI server demand surges on hyperscale and inference workloads, lifting CSP rack-scale buildout amid grid constraints.
NA CSPs raised Capex sharply, cementing AI infra as core strategy and powering AI server and data center growth.
Leveraging TrendForce’s extensive research on the AI supply chain, this report expands its focus to high-growth cloud service providers (CSPs). It offers comprehensive medium-to-long-term forecasts for AI servers, global data center deployment, and critical AI infrastructure—including power, thermal management, and networking.
NVIDIA's strategic investment in Lumentum and Coherent secures critical InP laser and high‑power CW light‑source capacity, shifting supply‑chain focus from module assembly to upstream optical components. With TSMC's silicon photonics platform and advanced OSAT capabilities, CPO is poised to play a central role in high‑power switches and high‑density AI racks at 800G/1.6T and beyond. Future CPO penetration will depend on light‑source stability and capex pace, serving as a vital engine for AI fabric scalability.
Google integrates TPU v7/v8, Ironwood racks, and Apollo OCS into a unified fabric, shifting the scaling unit from servers to racks. This drives 800G+ optical module penetration above 60% by 2026. Supply chain focus has shifted to laser chips and MEMS capacity. Strategic decisions must now track 800G/1.6T transceiver deployments, alongside GPU/TPU shipments, to accurately gauge the AI compute cycle.
Research Areas:
Industry News
