AI Chip Market 2026: Nvidia, AMD & Custom Silicon

Home / Indexes / Technical Performance / AI Chip Market 2026

Thank you for reading this post, don't forget to subscribe!

How big is the AI chip market in 2026?

Gartner forecasts worldwide semiconductor revenue to exceed $1.3 trillion in 2026, its strongest growth in two decades, with AI semiconductors accounting for roughly 30% of total revenue as hyperscaler infrastructure spending climbs more than 50% year over year (Gartner, April 8 2026). IDC pegs data center semiconductor revenue at $477.1 billion in 2026, with the intelligent datacenter segment of CPUs, GPUs, accelerators and custom ASICs now the largest slice of non-memory silicon (IDC, 2026 semiconductor forecast). On the systems side, TrendForce expects AI server shipments to grow 28.3% in 2026, driven by North American cloud providers and sovereign-AI demand (TrendForce, Jan 20 2026).

Nvidia’s dominance is still the story

Nvidia reported fiscal 2025 data center revenue of $115.2 billion, up 142% (Nvidia Q4 FY25 8-K, Feb 26 2025), then set a record $51.2 billion in a single quarter in Q3 FY26, up 66% year over year on Blackwell demand (Nvidia Q3 FY26 8-K, Nov 2025). Its grip on the merchant market remains near-total: Jon Peddie Research put Nvidia at 94% of add-in-board GPU shipments in Q4 2025, with AMD at 5% and Intel at 1% across 11.5 million units (JPR, Feb 2026). In AI servers specifically, TrendForce still sees GPU-based systems — overwhelmingly Nvidia — at 69.7% of 2026 shipments.

AMD builds a credible number two

AMD’s data center segment reached $16.6 billion in 2025, up 32% from $12.6 billion, led by Instinct MI350-series GPUs (AMD FY2025 results). The company has guided to a data center AI CAGR above 80% over the next three to five years and, in a landmark deal, agreed with OpenAI to deploy 6 gigawatts of AMD GPUs, beginning with a 1-GW rollout of MI450 in the second half of 2026 (AMD, Oct 2025). Management has publicly targeted roughly $20 billion in annual data center GPU revenue within two years (Reuters/Digitimes, Nov 2025).

Custom silicon is the fastest-growing front

Every major hyperscaler now ships its own accelerator — Google’s TPU (Ironwood/v7), Amazon’s Trainium3, Microsoft’s Maia 200, and Meta’s MTIA line. TrendForce expects custom ASIC systems to reach 27.8% of AI server shipments in 2026, growing 44.6% year over year against 16.1% for merchant GPUs. Broadcom, the dominant ASIC design partner, booked $20 billion of AI revenue in FY2025 with Q4 AI up 74%, and cited multi-billion-dollar XPU orders including an $11 billion follow-on for delivery into late 2026 (Broadcom Q4 FY2025 results, Dec 2025). Marvell posted $5.77 billion in fiscal-2025 revenue with AI near half of data center sales (Marvell FY2025 8-K).

Packaging and memory set the ceiling

Supply, not design, is the binding constraint. TSMC is scaling CoWoS advanced-packaging capacity toward roughly 130,000 wafers per month by the end of 2026, and TrendForce sees the CoWoS supply-demand gap narrowing from about 20% to 10% by year-end (TrendForce, June 15 2026). High-bandwidth memory is the other chokepoint: TrendForce expects HBM3E to make up about two-thirds of HBM shipments in 2026 as HBM4 ramps for Nvidia’s Rubin platform, with SK Hynix retaining the market lead ahead of Samsung and Micron.

Export controls redraw the map

US export rules continue to reshape demand. Nvidia took a $4.5 billion charge in Q1 FY26 tied to excess H20 inventory and purchase obligations after China-bound sales were curtailed (Nvidia Q1 FY26 10-Q). The company has stated it effectively cannot ship a competitive data center product into China under current rules, foreclosing a market that previously generated billions in annual revenue — a shift that is accelerating both domestic Chinese chip efforts and hyperscaler self-reliance elsewhere.

Frequently asked questions

What is Nvidia’s share of the AI chip market in 2026?

Nvidia remains dominant. TrendForce puts GPU-based AI servers — overwhelmingly Nvidia — at 69.7% of 2026 shipments, and Jon Peddie Research recorded 94% of add-in-board GPU units in Q4 2025.

How fast is custom AI silicon growing?

TrendForce forecasts custom ASIC AI-server shipments to rise 44.6% in 2026 — nearly triple the 16.1% growth of merchant GPUs — reaching about 27.8% of the market.

What is the biggest supply constraint?

Advanced packaging and memory. TSMC is expanding CoWoS toward ~130,000 wafers/month by end-2026, while HBM3E accounts for roughly two-thirds of 2026 HBM shipments as HBM4 ramps.

Sources

• Nvidia, Q4 FY2025 results (data center $115.2B, +142%), Feb 26 2025 — sec.gov
• Nvidia, Q3 FY2026 results (record data center $51.2B, +66%), Nov 2025 — sec.gov
• Nvidia, Q1 FY2026 10-Q ($4.5B H20 charge), 2025 — sec.gov
• AMD, FY2025 results (data center $16.6B, +32%; OpenAI 6-GW deal), 2025-26 — sec.gov
• Broadcom, Q4 & FY2025 results (AI revenue $20B; Q4 AI +74%; XPU orders), Dec 2025 — investors.broadcom.com
• Marvell, FY2025 results ($5.77B revenue; AI ~50% of data center) — sec.gov
• Gartner, Worldwide Semiconductor Revenue Forecast ($1.3T+ in 2026), Apr 8 2026 — gartner.com
• IDC, Semiconductor Market Forecast 2026 (data center semis $477.1B) — idc.com
• TrendForce, AI Server Shipments Forecast (+28.3% in 2026; GPU 69.7% / ASIC 27.8%), Jan 20 2026 — trendforce.com
• TrendForce, TSMC CoWoS supply-demand gap (20% to 10% by end-2026), Jun 15 2026 — trendforce.com
• Jon Peddie Research, Q4 2025 GPU shipments (Nvidia 94% AIB share) — jonpeddie.com