Arm predicts its chips will power 90% of AI servers by 2029, sidelining x86 and RISC‑V.

Arm’s momentum in the hyperscale market is now measurable in concrete deployment plans rather than speculative hype. Amazon Web Services, Google Cloud and Microsoft Azure have each announced that their next‑generation AI clusters will be built around proprietary Arm‑based CPUs, a shift that Counterpoint Research says will push Arm‑custom silicon to power 90 % of AI servers by 2029. The firm’s forecast, cited by Tom’s Hardware, is based on the rapid rollout of AWS’s Graviton processors paired with Trainium AI accelerators, Google’s Axion Arm CPU that underpins its upcoming TPU generation, and Microsoft’s Azure Cobalt CPU integrated with its Maia ASICs. By contrast, the legacy x86 ecosystem—still dominant in today’s data centers—will be relegated to a niche role, supplying only about 10 % of AI workloads, according to the same Counterpoint analysis.

The economics driving the transition are clear. Arm‑based custom CPUs are engineered for the data‑intensive patterns of modern machine‑learning models, delivering higher performance per watt and lower total‑cost‑of‑ownership than the general‑purpose Xeon and EPYC processors that have long powered enterprise servers. Counterpoint’s vice‑president of research, Neil Shah, notes that “backward compatibility with x86 is not vital” for emerging AI workloads, allowing hyperscalers to prioritize efficiency over legacy software constraints. This calculus is reinforced by the fact that the major cloud providers are already investing heavily in in‑house silicon design teams, reducing reliance on external chip vendors and capturing more of the value chain.

Meta’s involvement adds another layer of validation. The social‑media giant has become the “alpha customer” for Arm’s forthcoming AGI processor, a bespoke CPU that is expected to join the company’s internal AI infrastructure later this year. While Tom’s Hardware does not disclose deployment timelines, the partnership signals confidence that Arm’s ISA can support the most demanding generative‑AI workloads at scale. Even as some off‑the‑shelf x86 servers will continue to serve legacy applications, the bulk of future AI compute capacity is being earmarked for Arm‑centric designs, a trend that Counterpoint expects to accelerate markedly in the second half of 2026.

From a market‑share perspective, the shift reshapes the competitive landscape for Intel and AMD. Both firms have historically leveraged their x86 dominance to secure the majority of server contracts, but the forecasted 90 % Arm penetration implies a steep decline in new AI‑focused orders. Analysts at Counterpoint suggest that the incumbents will need to become “more flexible,” potentially by offering Arm‑compatible silicon or by deepening collaborations with cloud providers on custom solutions. So far, neither Intel nor AMD has announced a comparable roadmap to match the integrated Arm‑CPU/AI‑accelerator stacks that AWS, Google and Microsoft are deploying.

The broader implication for the semiconductor ecosystem is a rebalancing of design resources toward heterogeneous architectures. As hyperscalers roll out vertically integrated AI platforms—combining Arm CPUs, bespoke ASICs and specialized interconnects—the traditional separation between general‑purpose processors and AI accelerators blurs. This convergence could spur a wave of software innovation focused on Arm‑native toolchains, further entrenching the architecture’s foothold. Moreover, the projected dominance of Arm in AI servers aligns with the industry’s push toward energy‑efficient compute, a priority underscored by the increasing scrutiny of data‑center power consumption.

In sum, the convergence of cloud‑provider silicon strategies, cost‑efficiency imperatives and targeted performance gains underpins Counterpoint’s 90 % projection. If the rollout proceeds as outlined, Arm will not merely complement existing server architectures—it will become the de facto backbone of AI infrastructure, relegating x86 and even RISC‑V to peripheral roles in a market that is rapidly redefining the parameters of high‑performance computing.