A Futurum graphic of the CPU supply chain gives agentic AI its own row among the markets processors are sold into, and names six suppliers in it: Arm, NVIDIA Vera, Qualcomm, Intel, AMD and SiFive. By our count, three of the six are Arm-based under the graphic’s colour key.
The Agentic row is the shortest of the five end-market rows by our count, against 15 companies in Cloud, 10 in Enterprise, 8 in Consumer and 7 in Mobile. The graphic notes that “Companies may appear in multiple buckets. Not exhaustive.”
Business Pill · Lead Time Explained
A short explainer on lead times. It teaches the general idea only and says nothing about the companies in this piece.
The key insight: As we read Futurum’s graphic with its February release, agentic AI has become a market that CPU makers are sold into on its own, with six suppliers named in that row and three of them Arm-based by our count.
The graphic, titled “CPU Supply Chain”, has three panels: Design, Manufacturing and CPUs by end market. Design lists the instruction sets: x86 from AMD and Intel, Arm, RISC-V from Andes, SiFive, MIPS, Nuclei and XuanTie, and proprietary instruction sets from IBM and Loongson. EDA tools come from Cadence, Siemens and Synopsys.
Manufacturing lists wafer fab equipment from ASML, TEL, Applied Materials, KLA, Lam Research and ASM, and foundries TSMC, Intel, Samsung, GlobalFoundries, UMC, SMIC and Rapidus. Memory comes from Samsung, SK hynix and Micron, alongside packaging, substrate, DPU and test suppliers.
The end-market panel colours each company by instruction set. By our reading of that key, the Agentic row holds Arm, NVIDIA Vera and Qualcomm on Arm, Intel and AMD on x86, and SiFive on RISC-V. The graphic shows no date.
Companies named per end-market row in Futurum’s CPU Supply Chain graphic, FourWeekMBA’s count: Agentic 6, Cloud 15, Enterprise 10, Consumer 8, Mobile 7; the Agentic row by the graphic’s colour key: Arm 3 (Arm, NVIDIA Vera, Qualcomm), x86 2 (Intel, AMD), RISC-V 1 (SiFive). Futurum notes the map is not exhaustive.
Why Agentic Gets Its Own Row
Futurum’s analyst Brendan Burke made the case in a 24 February press release: “a quiet supply crisis has emerged in the CPU market, driven not by traditional server refresh cycles, but by the explosive growth of agentic AI and reinforcement learning workloads”.
The release says these workloads are “pushing CPU-to-GPU ratios in AI clusters back toward 1:1”, and that “Intel and AMD confirming high-core-count server processors are effectively sold out”. It adds that NVIDIA’s “standalone launch of its Vera CPU in Q1 2026” came with “a massive multi-year deal with Meta, which includes the first large-scale deployment of standalone NVIDIA CPUs to support the operation of personal agents”.
On RISC-V, the release says it “is becoming a viable high-performance contender through a strategic partnership between SiFive and NVIDIA”. That is the one RISC-V name in the graphic’s Agentic row.
What Agent CPU Demand Looks Like in Numbers
A UBS model of Meta’s Muse agent, as summarised on X by Shay Boloor, shows how the count of server chips depends on how many agents run at once. The post says UBS puts the base case at “~3,000 chips if 750K agents run simultaneously with 4x CPU sharing” at 3 million daily users.
The same post gives a high-engagement case of “~23,800 chips if 1.5M agents run simultaneously without CPU sharing”, and says the same base assumptions “would imply nearly 100,000 EPYC chips at 100M daily users”. By our arithmetic, the high case is about eight times the base case.
The Business Lever: The Orchestration Layer
As we read the graphic with Futurum’s release, the Agentic row marks CPUs moving from host processors to the layer that schedules and runs agents. The release puts it as “The traditional host processor role is dead”. On the GPU side, our Rubin benchmark piece covered the accelerators those CPUs sit beside.
The Structural Read
As we read the graphic, the Agentic row is a narrower field than Cloud: 6 companies against 15, by our count.
By our count under the graphic’s colour key, the Agentic row splits three Arm (Arm, NVIDIA Vera, Qualcomm), two x86 (Intel, AMD) and one RISC-V (SiFive).
By our arithmetic on the UBS figures shared on X, the high-engagement case for Meta’s Muse needs about eight times the server chips of the base case at the same 3 million daily users.
Futurum, Brendan Burke, 24 February 2026
“The traditional host processor role is dead.”
Three Implications
FOR CHIP SUPPLIERS Futurum’s release says high-core-count server processors from Intel and AMD are effectively sold out and that the CPU market could exceed GPU and XPU growth by 2028.
FOR HYPERSCALERS The release says Google has migrated 30% of its internal applications to its custom Axion chips and that Microsoft’s Cobalt 200 is scaling for newer AI workloads.
FOR AGENT BUILDERS The UBS figures shared on X show the chip count turning on how many agents run at once and how much CPU capacity is shared: about 3,000 chips in the base case and about 23,800 in the high case.
The graphic shows no date, and we could not find the Futurum report it comes from; it was passed to us as an image. The company counts are our own from the graphic, and Futurum says the map is not exhaustive and companies can appear in several rows. The UBS figures come from an X post summarising the model, not from the UBS note itself. The eight-times ratio is our arithmetic.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Futurum’s CPU supply-chain graphic gives agentic AI its own end-market row with six suppliers, three of them Arm-based by our count, against 15 in Cloud. Its February release says agentic and reinforcement-learning workloads are pushing CPU-to-GPU ratios back toward 1:1, and a UBS model shared on X puts Meta’s Muse at about 3,000 to 23,800 server chips at 3 million daily users.
A note on sourcing. The company lists and counts come from a Futurum graphic passed to us as an image, which we redrew rather than republishing; the counts are ours. Futurum’s quotes come from its 24 February 2026 press release. The UBS figures come from an X post summarising the model. Comparisons labelled as our arithmetic are ours.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
