The server GPU market is driving the growth, reaching $101 billion in 2024, and is expected to more than double by 2030. With a  90%+ market share, Nvidia remains the dominant player, driven by the CUDA ecosystem, rack-scale solutions and scale-up technology, leveraging its full-stack approach. Its Blackwell platform, released in 2025, along with the upcoming Rubin platform expected in 2026, are poised to extend this lead.

AI ASICs peaked at $9 billion in 2024 and are expected to see double-digit annual growth through 2030, making them the second-largest contributor to the processor landscape.  AI ASICs are favoured by hyperscalers for cost-effective, high-efficiency inference, and are tightly co-developed with partners such as Broadcom, Marvell, GUC, Alchip, and MediaTek.

Custom silicon is being developed by hyperscalers  with Google, AWS, and Huawei investing in in-house chips (TPUs, Trainium, Ascend).

The server CPU market will reach $35.6 billion by 2030, with Intel and AMD facing increasing pressure from Arm-based alternatives. The CPU server market, led by Intel and AMD, which together hold 80% market share, remains essential for general-purpose compute but is under pressure.

Arm-based CPUs from hyperscalers and new players are gaining momentum, notably Amazon’s Graviton, Google’s Axion, and Nvidia’s Grace, with claimed major power-efficiency advantages.

Strategic sovereignty has emerged as a market driver, with countries investing in local AI compute infrastructure amid export controls and rising geopolitical tensions.

As AI becomes an asset in global digital strategy, governments are investing in dedicated AI datacentres to ensure national compute capability. In parallel, the U.S. government continues to enforce strict export controls, which divide the world into regulatory tiers and limit China’s access to cutting-edge AI chips.

In response, the Chinese authorities are accelerating China’s domestic semiconductor efforts, while Nvidia develops export-compliant chips. Meanwhile, Huawei is ramping up its CPU and AI ASIC development, underlining the strategic urgency around self-sufficiency in AI compute.

“Strategic compute becomes the heart of AI infrastructure,” says Yole’s Adrien Sanchez, “the rise of Generative AI is redefining compute architectures, our research has identified a shift away from general-purpose processors toward custom, AI-optimised silicon, across GPUs, ASICs, CPUs, DPUs, and more.”

“Hyperscalers are developing their own AI ASICs because Nvidia’s margin on server GPUs is over 85%, something never seen before in the processor industry,” says Yole’s Hugo Antoine, “with AI ASICs, they’re able to bring their costs down by an order of magnitude.”

Smaller segments are also evolving:

Server FPGAs: $1.5 billion by 2030,

DPUs & network ASICs: $17.7 billion by 2030,

Crypto ASICs: $4.2 billion by 2030, influenced by shifts in mining dynamics.

Yole’s analysts point to consolidation and mergers and acquisitions as key enablers of this compute evolution. 2024 saw:

SoftBank acquire Graphcore,

AWS invest $700M in Tenstorrent,

Rebellions and Sapeon merge in South Korea,

Meta attempt but fail to acquire Furiosa for $800 million.

These developments highlight the scarcity of competitive AI chip teams and the rising value of silicon expertise in AI infrastructure.