AI Chip Market to Reach $677.59 Billion by 2035 as Inference and Custom Silicon Reshape Global Compute Demand

DC Market Insights forecasts the global AI chip market will grow from $185.26 billion in 2025 to $677.59 billion by 2035, a 13.85% CAGR. It projects AI ASICs and SoCs to grow faster than GPUs—19.61% versus 11.96% annually—while inference spending reaches $439.05 billion by 2035. The report estimates NVIDIA held 72% of 2025 market value and highlights large AMD–OpenAI and Broadcom–OpenAI accelerator agreements; these are research forecasts and cited commitments, not reported market-wide results.

Analysis

The investable signal is mix shift, not the headline TAM: more inference and customer-specific silicon could widen the market while redistributing profits away from merchant GPUs. Broadcom may gain if custom-accelerator programs convert into shipped, profitable products; hyperscalers may capture lower inference costs, but higher utilization can also expand total compute demand. NVIDIA’s defense is not just GPU share: software compatibility, integrated systems, networking and supply availability can preserve pricing power even as ASIC share rises. The report does not establish that ASIC growth is incremental rather than substitution, or that announced capacity becomes funded, powered deployments. Treat its forecasts and supplier-share estimates as third-party scenarios, not company guidance.

Near term, the market may trade the optimistic projections; over the next 1–3 months, verify orders, customer capex and shipment timing in company disclosures. Across 6–18 months, power, advanced packaging and high-bandwidth memory availability—not chip demand alone—could set revenue conversion and shift value to constrained suppliers. The contrarian risk is that inference efficiency reduces compute cost per task faster than usage expands, while power delays strand planned capacity. Conversely, efficient inference may broaden use cases and increase total workloads. No valuation or consensus data here supports an outright directional call.

Key Decisions for Investors

• Do not buy the sector solely on this market forecast. Use upcoming earnings and disclosures to check hyperscaler capex, accelerator shipments, customer concentration and whether announced programs are funded and on schedule.
• Conditional relative-value idea: consider long AVGO versus short NVDA only if Broadcom reports converting custom-accelerator programs into recurring shipments while NVIDIA demand and system economics remain resilient. The thesis is mix-share gains, not an assumed collapse in GPU demand; abandon it if program timing slips or NVIDIA retains orders and pricing better than expected.
• Track NVIDIA and AMD shipment cadence alongside packaging, memory and power constraints. If deployment delays rise despite customer demand, favor infrastructure bottleneck beneficiaries only after confirming order conversion; otherwise avoid extrapolating chip TAM into near-term revenue.
• Falsifiers: material reductions in hyperscaler capex or accelerator guidance, delayed custom-silicon deployments, evidence that inference cost savings are reducing aggregate compute demand, or worsening power/packaging constraints that push shipments beyond planned windows.