Connected views of AI infrastructure supply, demand, forecasts, and market signals.
This dataset is a worldwide model of the server-class CPU market from 1994 through 2032. History through 2025 closes at $29.3 billion, 24.3 million sockets, and $1,207 a socket. Bear, Base, and Bull then separate, ending 2032 at $46.2 billion, $107.9 billion, and $161.8 billion.
This dataset is the chip and board layer of the data-center accelerator market from 1994 through 2032, covering merchant GPUs, merchant and captive AI ASICs, data-center FPGAs, and DPUs. The 2025 market history is $219.2 billion. Base 2032 is $1.35 trillion, with Bear at $550 billion and Bull at $2.40 trillion.
This dataset is data-center DRAM from 1994 through 2032: server modules, high-bandwidth memory sold into accelerators, CXL memory, and a small SOCAMM and custom-module line. The 2025 data-center total is $72.6 billion. Base 2032 is $166.7 billion, with Upside at $214.5 billion and Downside at $112.2 billion.
This dataset is the enterprise SSD market from 1994 through 2032, counting NAND drives sold into servers and storage systems. The 2025 market history is $27.81 billion, and the Base path for 2026 is $149.8 billion. Base 2032 is $167.7 billion, Bear is $85.7 billion, and Bull is $287.2 billion.
This dataset is the data-center networking market from 1994 through 2032: Ethernet switch systems, InfiniBand switches, the scale-up fabric, host adapters, and Ethernet optical modules. The 2025 total is $80.88 billion. Base 2032 is $236 billion, Bear is $140 billion, and Bull is $400 billion.
This dataset covers the server power supply and the rack PDU from 2025 through 2032. The 2025 actual is $8.63 billion. Base 2032 is $17.52 billion, Bear is $13.87 billion, and Bull is $21.30 billion. Cooling, UPS plants, and the utility feed are outside the total.
This dataset is a worldwide model of neocloud and Tier-2 GPU rental from 2023 through 2032. It counts rented accelerator capacity sold by specialized GPU clouds, and leaves out AWS, Azure, GCP, Oracle, chatbot subscriptions, and captive lab clusters. The 2025 market history is $12 billion. Base 2032 is $90 billion, with Bear at $47.9 billion and Bull at $129 billion.
This dataset is a worldwide model of the general-purpose humanoid robot market from 2024 through 2032. The 2025 close is $529 million of robot revenue, 18,000 robots shipped, and about $29,400 a robot. Base 2032 is $23.9 billion, with Conservative at $8.6 billion and Aggressive at $48.3 billion.
Three server bills of materials: hyperscaler-driven spend, enterprise OEM vendor revenue, and the combined total with each server counted once.
This dataset is the cash capital expenditure of the companies in the buyer-side model. The long file, Hyperscaler IT Spend From 1994, runs from 1994 through 2032. The years through 2025 are filings. From 2026 the same columns carry the forecast already used in the forecast workbook, and a filed year replaces a forecast. The thing being counted is purchases of property and equipment, on the fiscal year each company reports. Server spend and the bill of materials start in 1998, in the forecast workbook, once an IT share is applied.
This dataset is worldwide server vendor revenue, the supply side of the same machines the buyer model counts from capex. The long file, Server OEM Demand From 1994, runs from 1994 through 2032. 2025 is the last closed year in the existing model: $444.1 billion. From 2026 the market is the IDC vintage already used in the forecast workbook, then a deceleration through 2032. The thing being counted is what server vendors invoice. ODM Direct is IDC’s bucket for the unbranded channel, not the sum of the contract manufacturers’ own revenue.
This dataset puts the two long series next to each other. Total IT and Server Spend From 1994 runs from 1994 through 2032. Server vendor revenue is the supply side. Buyer fiscal capex is the cash the named companies spent. They are different objects, and the chart is there so a reader can see them diverge. The combined identity in the working total workbook — servers counted once, then IT and the eight components — still starts in 2015 and is not recomputed in the long file.
Signals connect component availability with delivery conditions and the capital spending that funds AI infrastructure demand.
Compares supply with demand for CPU, accelerators, RAM, and enterprise SSD. Values below 100% indicate a tighter market; values above 100% indicate that modeled supply is ahead of demand.
Tracks modeled delivery time for CPU, GPU, and HBM against market survey bands. Longer lead times generally indicate tighter availability, allocation, and stronger vendor pricing power.
Follows capital expenditure by major cloud providers, broader MANGOS companies, and neoclouds. These investments fund server, accelerator, memory, networking, and storage demand.