BofA 2030 Forecast: ARM Overtakes x86 in Server CPUs as Custom Silicon Erodes Nvidia's AI Accelerator Dominance

Server CPU Market: ARM's Rise and x86's Decline

Bank of America's latest research reveals a fundamental shift in the server CPU landscape. The total addressable market (TAM) is projected to expand from approximately $35 billion in 2025 to $210.6 billion in 2030, demonstrating robust growth despite AI accelerators capturing headlines.

Intel's value share is forecast to plummet from 40.4% in 2025 to 22.0% in 2030. AMD grows steadily from 27.4% to 30.7%, securing the second position. The ARM ecosystem emerges as the largest winner, with two distinct segments:

ARM Merchant (commercial ARM) : Share surges from 29.3% to 37.9%, becoming the single largest segment. This is driven by cloud providers' pursuit of performance-per-watt, exemplified by AWS Graviton and Ampere processors.

ARM Custom (in-house ARM) : Share rises from 2.7% to 9.4%, reflecting hyperscalers' trend toward custom silicon optimized for specific workloads.

Combined ARM share reaches 47.3% by 2030, signaling a transition from "general-purpose compute first" to "efficiency and customization first" in data center architecture.

AI Accelerator (XPU) Market: From Monopoly to Multi-Polar Competition

The AI accelerator market is measured in gigawatts (GW) of deployed capacity — a more physically meaningful metric than unit shipments. Annual deployment capacity is projected to grow from 13.0 GW in 2025 to 55.7 GW in 2030, a 4.3x increase, indicating that energy supply will become the core bottleneck for AI infrastructure scaling.

Nvidia's absolute deployed capacity continues rising (from ~10 GW to ~25 GW), but its share of total capacity is significantly diluted. In 2025 Nvidia holds near-total dominance; by 2030 the landscape fragments into:

Hyperscaler custom ASICs : Google (TPU), AWS (Trainium), Meta (MTIA), and Microsoft (Maia) show substantial growth in the 2026–2030 period, reflecting large-scale deployment of in-house chips to reduce Nvidia dependency and optimize total cost of ownership (TCO).

New entrants : OpenAI appears as a distinct segment, suggesting chip design or massive cluster initiatives; Anthropic's presence via TPU highlights deep model-company–chip-vendor coupling.

AMD : Grows steadily but remains smaller than Nvidia.

The market shifts from "one dominant player" to "one superpower with multiple strong contenders." Nvidia retains leadership via CUDA ecosystem and rapid hardware cadence, yet a hybrid deployment model — custom ASICs alongside merchant GPUs — becomes standard for hyperscale data centers.

Key Conclusions

Heterogeneous compute architectures : Future data centers will not be x86 monocultures. At the CPU layer, ARM claims half the market; at the accelerator layer, GPUs coexist with diverse custom ASICs (TPUs, etc.).

Performance-per-watt is the core metric : Both ARM's CPU ascendancy and custom XPU proliferation are driven by energy efficiency. At GW-scale deployments, performance-per-watt directly determines operational cost and expansion limits.

Vertical supply-chain integration : Large cloud and AI firms are integrating upstream through custom silicon (ARM Custom + custom XPUs) to build moats beyond commodity hardware.

By 2030 the compute market will be six times larger, but its underlying logic has changed: the era of general-purpose hardware dividends has ended; the age of customization and efficiency optimization has fully arrived.