Evergreen Capital
Technology Investor, Semiconductors & AI Infrastructure
Google's TPU v8 design is very positive for Arm because it puts more Arm CPUs into every AI inference server. For the first time, Google's single-threaded Arm-based Axion CPU serves as the TPU head node. To meet surging agentic demand, Google is also doubling the number of physical CPU hosts per TPU inference server, raising the CPU-to-accelerator ratio. x.comÂ
Arm looks set to run away with the CPU market as CPUs become the bottleneck for agentic AI workloads. It reached about 25% server CPU share by end-2025, up from 15%, per Dell'Oro, on better price-performance and efficiency than x86, and powers over 50% of new AWS CPU capacity via Graviton. Agentic AI drives rising batch sizes that worsen the CPU bottleneck, where the Arm AGI CPU's single-threaded design looks structurally advantaged over x86 and possibly Nvidia's Vera. x.comÂ
Arm is shifting from secondary supplier to the core orchestration backbone of AI data centers. Agentic AI needs roughly four times as many CPU cores per gigawatt, part of a $100 billion CPU market Arm sees for 2026. Its new AGI CPU line has over $2 billion in secured demand, with Oracle Cloud and ByteDance joining, and Google will use Arm-based Axion CPUs in all next-gen TPU systems. Hyperscalers' custom chips standardize on Arm's Neoverse, and Armv9 carries about double v8's royalty rate. x.comÂ
Google TPU adoption could broaden the AI hardware opportunity beyond Nvidia, with demand reaching custom ASICs, HBM, packaging, optics and testing. Per The Information, OpenAI reportedly plans to buy Google's TPUs and is hiring compiler engineers, but no purchase is confirmed. Broadcom, SK hynix, Samsung, Micron, TSMC, Coherent, Lumentum, FormFactor and Teradyne could benefit. x.comÂ
Arm cannot even raise its $2 billion customer-interest figure in what should be the greatest CPU market of our lifetimes, and whether it can deliver even that is still unproven. x.comÂ
Google's allocation of massive TPU capacity to Anthropic strains its own AI effort while its researchers face compute constraints. CNBC reported internal tension; Anthropic competes with Gemini. Google should keep more TPUs for its own AI and buy more NVIDIA GPUs to rent to cloud customers, serving CUDA developers alongside Gemini. Selling TPU systems changes the equation, but Alphabet does not disclose TPU margins, so NVIDIA-like economics cannot be assumed. x.comÂ
Broader adoption of custom AI silicon could benefit Broadcom and Marvell. OpenAI reportedly plans to expand its use of Google's TPUs beyond Nvidia GPUs, and Google TPUs are coming to Oracle Cloud in 2027. Both are potential tailwinds for custom-chip suppliers. x.comÂ
Google's TPUv7 inference efficiency supports the bullish case for Alphabet. Per SemiAnalysis' first TPUv7 FP8 vLLM benchmark submission, Google's TPUs delivered 67% more AI inference performance per dollar than Nvidia's GB300 NVL72, even with Nvidia's system using FP4 precision. x.com x.comÂ
Alphabet's ownership of the whole AI stack, from chips to cloud to distribution, gives it a structural cost advantage rivals can't easily copy. It alone designs its own TPUs, runs a global cloud and owns Search, Android, YouTube and Workspace, so aggressive inference pricing is a strategic weapon. As inference overtakes training, case studies show moving workloads from GPUs to TPUs can cut operating costs by over 50%. Microsoft and Amazon are building custom chips, but Alphabet is years ahead. … Show more
Arm is no longer just a royalty collector, a radical change in its business, and agentic AI demand for sold-out CPUs is amplifying the move. Arm-based CPUs could capture nearly 90% of host CPU deployments in custom AI ASIC servers by 2029, and Nvidia was expected to unveil a new Arm-based processor at Computex. Not all of the rally is fundamentals, and the stock could pause around $400; in this kind of market, price action works better than P/E. x.com x.comÂ
Arm is positioned to win from a CPU shortage that x86 cannot meet. Intel and AMD raised prices 10-15% and lead times stretched from 1-2 weeks to six months, as agentic AI drives CPU demand. Arm delivers up to 60% better performance per watt than x86 in cloud workloads, and ASUS says 30% of its Copilot AI PCs now use Arm, up from 20%. Arm also launched its own 136-core AGI CPU with Meta as lead partner plus OpenAI and Cloudflare, expecting $15 billion in annual revenue within five years. x.com … Show more
Arm is well positioned in the AI CPU stack as agentic AI drives massive CPU demand and x86 loses ground. Its new AGI CPU, its first production silicon, comes with a five-year goal of about $15 billion in chip revenue, $25 billion total and more than $9 in EPS, roughly 38% annual growth, though meaningful revenue is not expected until fiscal 2028. Royalties from third-party CPUs like Nvidia's Vera add another path. That TAM expansion could take out its roughly $185 all-time high within two years. … Show more
