DEV Community

AtlasPCBEngineering
AtlasPCBEngineering

Posted on • Originally published at atlaspcb.com

Citi Raises AI-PCB Market Forecast to ¥562B by 2028: Google TPU to Surpass Nvidia

Citi's New AI-PCB Market Forecast (June 21, 2026)

Citibank just released a major update to its AI-PCB market analysis, significantly raising total addressable market forecasts:

Year TAM (RMB) YoY Growth Key Driver
2026 152 billion 86% GPU demand surge, initial ASIC ramp
2027 307 billion 102% ASIC dominance, optical module explosion
2028 562 billion 83% Google TPU scale, network infrastructure

The report is notable for formally projecting Google TPU PCB procurement exceeding Nvidia GPU-related demand by 2028.


Demand Breakdown: No Longer Just "Nvidia GPU Boards"

The 2027 demand composition reveals a diversifying market:

  • ASICs (custom silicon): 34% — Google TPU, Amazon Trainium, Microsoft Maia
  • Nvidia GPU-related: 24% — still significant but no longer dominant
  • CPU interconnect: 16% — expanding I/O on high-core-count server CPUs
  • Switches: 14% — 51.2T/102.4T switching silicon
  • Optical modules: 12% — 1.6T optics requiring multi-layer substrates

This diversification matters because each category has different PCB requirements:

  • GPU carriers: heavy copper (4-5oz), high layer count (20+), power delivery focus
  • ASIC boards: vary by customer, generally similar to GPU
  • Optical module substrates: thin profile, tight impedance, low-loss materials
  • Switch boards: extreme signal integrity on 56+ Gbps SerDes

Google TPU: The New PCB Demand Leader by 2028

Citi expects Google TPU PCB procurement to reach $16 billion by 2028, surpassing Nvidia GPU-related demand. Between 2027-2028 incremental demand:

  • Google TPU: ~30% of new demand
  • Nvidia: ~25%
  • Switches + optical: ~28% combined

Unlike Nvidia's distributed ecosystem (multiple OEM partners), Google's procurement is centralized — creating massive order volumes for qualified suppliers but also concentration risk.


Optical Module PCBs: Fastest-Growing Segment

The standout growth number: 135% growth in 2026, 178% in 2027 for optical module PCBs.

Optical module substrates have different requirements than compute boards:

  • Thin profile: 0.8-1.2mm
  • Moderate layer count: 8-14 layers
  • Tight impedance: ±5% on 50GHz+ channels
  • Low-loss materials: Dk < 3.0, Df < 0.002 at 28GHz
  • Fine trace/space: 3/3mil for 112G PAM4

A manufacturer optimized for thick, heavy-copper GPU boards may miss this growth without process investment.


The Capacity Constraint Warning

The most actionable element for hardware engineers:

  • Chinese PCB manufacturers need 13-15 months from greenfield to production
  • Upstream CCL and fiberglass suppliers need 18+ months for capacity additions
  • Citi expects next expansion announcements in H2 2026 for 2028 demand
  • AI-CCL supply may become tighter than fabrication capacity

What this means for you:

If your board requires HDI, heavy copper, or high-layer-count fabrication with production timeline into 2027:

  1. Secure manufacturing capacity commitments early
  2. Expect lead times extending from 3-4 weeks to potentially 6-8 weeks
  3. Specify standard materials where possible — exotic laminates face tightest supply
  4. Source secondary vendors for production redundancy

Practical Takeaways for PCB Buyers

The report confirms several trends we've observed in our facility:

  1. Material procurement is becoming the bottleneck — not fabrication capacity. Ultra-low-loss laminates and specialty prepregs face the tightest supply.

  2. Diversify your supplier base — concentration risk increases as fewer manufacturers can handle the full range of AI-PCB requirements (heavy copper + fine line + low-loss materials).

  3. Lock in lead times now — if your production timeline extends into 2027, current-day lead time quotes may not hold as capacity fills.

  4. Design for manufacturability — boards using standard, widely-available materials (Shengyi S1000-2M, ITEQ IT-180A) will face shorter lead times than those specifying exotic laminates.


For detailed analysis including supply chain implications and material sourcing strategies, see: Full Analysis on AtlasPCB


Building AI hardware? We fabricate 16-30 layer boards for accelerators with dedicated capacity allocations. Get capacity-guaranteed pricing →

Top comments (0)