Data Center Accelerator Market to Reach USD 605.42 Billion by 2035 as Rack-Scale AI Systems and Inference Demand Accelerate Deployment

DC Market Insights estimates the data center accelerator market grew to USD 169.77 billion in 2025 and forecasts USD 605.42 billion by 2035, a 13.56% CAGR. Rack-scale systems are projected to account for 61.8% of market value by 2035, amid expanding AI infrastructure spending and demand for integrated platforms; inference workloads are also creating room for custom accelerators. North America held 59.4% of the market in 2025, while export controls and geopolitical restrictions remain constraints in selected markets.

Analysis

The investable signal is a change in value capture, not the market-size forecast. Rack integration can increase the dollars attached to each deployment while shifting economics toward networking, memory, power delivery and cooling; a higher rack-scale share does not automatically mean higher chip-vendor margins. Inference also makes accelerator demand more substitutable: custom silicon can take workloads where cost per token matters, but broad demand growth need not be zero-sum if total compute expands.

Potential relative beneficiaries are Broadcom (AVGO) and Marvell Technology (MRVL) through custom silicon and connectivity, alongside incumbent platforms such as NVIDIA (NVDA) that can sell integrated systems. AMD (AMD) benefits if announced deployments convert into shipped, production-ready capacity. The main second-order constraint is power and data-center construction: delayed energization could defer accelerator revenue across vendors and strand planned capacity, while export controls can redirect—not necessarily eliminate—supplier demand.

Treat the release as a vendor-sponsored market estimate, not a near-term earnings signal. Its long-dated forecast and large deployment announcements are not independently established revenue; timing, customer concentration, utilization and delivered systems matter more than headline capacity. Over days, sentiment may favor AI infrastructure; over 1–3 months, verify reported orders and shipment timing; over 6–18 months, watch inference mix, custom-ASIC share and power availability. The contrarian risk is that investors extrapolate rack value growth into durable supplier pricing power even as custom designs and system integration pressure chip ASPs.

Key Decisions for Investors

• Do not trade the 2035 market forecast as an earnings catalyst. For the next 1–3 months, use AMD, AVGO and MRVL as a watchlist for evidence of shipped capacity, customer concentration and revenue conversion; treat announced capacity or agreements alone as insufficient confirmation.
• Consider a small relative-value position long AVGO/MRVL versus underweight NVDA only if customer disclosures show custom-accelerator deployments scaling and NVDA order growth or pricing evidence weakening. This is a share-shift thesis, not a call that total AI compute demand contracts; exit if custom deployments slip or NVDA maintains strong order conversion and platform economics.
• For a less directional expression, favor diversified exposure across accelerator and connectivity suppliers over a single-chip thesis, while monitoring data-center power and construction milestones. A broad delay in energization would threaten revenue timing for the entire supply chain, not just one vendor.
• Falsifiers: major customer deployment delays, weaker accelerator-related guidance, evidence that custom ASICs remain limited to narrow workloads, or power constraints that push planned capacity beyond the next 6–18 months. Verify actual shipments, utilization and customer capex before increasing exposure.