Compute & silicon·Updated May 29, 2026

Custom silicon (hyperscaler ASICs)

Hyperscalers are designing their way around Nvidia with bespoke accelerators — but only two firms can co-design them at scale, and they all funnel through the same TSMC nodes and CoWoS slots.

Tightness gaugeCritical

Tightness

75/ 100Critical

+4 this quarter — ASIC shipments now outgrowing merchant GPUs for the first time, with new programs (Apple, OpenAI, Anthropic) added — but the design duopoly and shared TSMC/CoWoS chokepoints keep it tight.

Coverage35%
70
Lead time25%
70
Concentration20%
88
Momentum20%
75
2026 ASIC shipment growth
+44.6%YoY vs +16% for GPUs
as of 2026-05-26·TrendForce via Tom's Hardware
Co-design market share
~95%Broadcom + Marvell
as of 2026-05-26·TechTimes
Broadcom AI backlog
$73B→ $100B target 2027
as of 2026-05-19·Tom's Hardware
ASIC share of AI servers
~27.8%2026E, highest since 2023
as of 2026-05-26·TrendForce via Tom's Hardware
Why it gates the buildout

Hyperscaler accelerators — Google TPU, Amazon Trainium, Meta MTIA, Microsoft Maia — are how the largest buyers cut their Nvidia dependence. But the hyperscalers don't build them alone: Broadcom and Marvell together do ~95% of the custom-ASIC co-design work, and every one of these chips is fabricated by TSMC on advanced nodes and assembled with CoWoS. So custom silicon doesn't escape the compute bottleneck — it inherits the same foundry and packaging chokepoints as merchant GPUs.

Who's exposed
CompanyRoleExposure
~70%+ of custom-AI co-design; TPU/MTIA; six XPU customersDirect
~20–25% share; Trainium, Maia; $75B design pipelineDirect
AlchipTPE 3661
Tier-2 ASIC design/back-end houseHigh
MediaTekTPE 2454
Entering custom AI; ~$2B Q4'26 programWatch
TSMCTSM
Fabricates every hyperscaler ASIC — shared chokepointHigh
Catalysts & timeline
Q4 2026MediaTek's first hyperscaler ASIC ramps; ~$2B revenue, double prior guidance
2027Broadcom targets $100B AI chip revenue (timing may slip to late 2027/2028 on TSMC capacity)
2026–27OpenAI, Anthropic, Apple custom programs move toward volume
What would loosen it

A generational efficiency leap from Nvidia or AMD that undercuts the economics of building custom chips, or TSMC advanced-node and CoWoS capacity catching up. Neither looks likely to bite before mid-2027, which is when TSMC constraints are expected to ease.

Latest developments
2026-05-26
TrendForce: custom ASIC shipments to grow 44.6% in 2026 vs 16.1% for merchant GPUs
2026-04-30
MediaTek doubles AI-ASIC guidance to ~$2B for Q4 2026; second program targeting 2027 mass production
2026-03-12
Broadcom Q1 FY26 AI revenue $8.4B (+106% YoY); confirms six XPU customers including Apple, OpenAI, Anthropic
Linked bottlenecks
Leading-edge foundry (2nm / N2)CoWoS advanced packagingHBM (high-bandwidth memory)Silicon interposers

Get the squeeze before the headlines.

A weekly read on the AI buildout's tightest constraints. Free, sourced.

Subscribe