hardware.co · AI hardware for on-prem LLM deploymentsVendor-neutral · Quote onlyBuilt to spec
Graphic · AI Compute · 2025–2026

AI Company Costs Breakdown: Compute vs. Staff (2025–2026)

The Compute Ruler: line up the filed accounts of Anthropic, MiniMax and Z.ai and compute lands in the same place every time, at 54.5% to 63.8% of operating costs. Staff, the famous AI salaries included, took 14.1% to 24.0%.

Updated Nov 17, 20267 charts · 4 key stats5 primary sources
Fig. 1

The Compute Ruler

AI company expenses broken down by what the money bought, from each company's own filings. Compute is drawn flush left, so its share lines up like a mark on a ruler: more than half of operating costs at every company, every period.

ComputeStaffMarketing & partnershipsOther
0%25%50%75%100%Compute share: 54.5%–63.8%Anthropic, 2025US$ thousandsFull year 2025Compute 57.9%Staff 14.5%4.5%Other 23%Anthropic, 2026 YTDUS$ thousandsJan–Sep 2026Compute 54.5%Staff 14.1%16%Other 15.4%MiniMaxUS$ thousandsJan–Sep 2025Compute 63.8%Staff 17.9%12.8%5.5%Z.ai (Zhipu)RMB thousandsJan–Jun 2025Compute 57.7%Staff 24%4.6%Other 13.8%

Operating costs are cost of revenue (or sales) plus operating expenses, including stock-based pay. Non-operating items such as fair-value changes on convertible notes and preferred shares are excluded. Anthropic reports its costs by nature; the MiniMax and Z.ai splits are our sums of the line items in each prospectus. Values are in thousands; Anthropic reports in millions. Periods differ, and Z.ai is in RMB, so compare shares, not amounts.

Anthropic's compute and infrastructure costs, 2025
$7,326M
57.9% of its $12,645M operating expenses (derived)
Compute's share of operating costs at Anthropic, MiniMax and Z.ai
54.5%–63.8%
Latest disclosed periods, 2025–2026 (derived from filings)
Anthropic's compute costs as a share of revenue
251% → 58%
Q1 2025 to Q3 2026, as reported in the S-1
Anthropic's committed future spend, mostly compute
$900.9B
$451.6B of it expected before the end of 2030
Fig. 2

Compute alone cost Anthropic about 1.6 times its 2025 revenue

Anthropic revenue and operating expenses, 2025, US$ millions

Revenue$4,585MCompute & infra$7,326MPersonnel$1,839MBartz settlement$1,500MOther & admin$1,412MSales & marketing$568M

Full S-1 line names: compute and infrastructure; personnel; Bartz settlement; other operating and administrative; sales, marketing, and partnerships. Compute and infrastructure covers cloud compute fees, storage, networking, colocation, equipment leases and committed power. The multiple is $7,326M ÷ $4,585M = 1.598. Total operating expenses were $12,645M, for an operating loss of $8,060M.

Fig. 3

Anthropic's revenue overtook its compute bill in early 2026

Anthropic quarterly revenue vs. compute and infrastructure costs, US$ millions

0$6B$12B$18BQ1 2025Q3 2025Q1 2026Q3 2026Revenue$16,956M in Q3 2026Compute$9,863M in Q3 2026

Quarterly figures are unaudited. Compute and infrastructure costs rose in every quarter, but revenue rose faster: the S-1 puts compute at 251% of revenue in Q1 2025 and 58% in Q3 2026.

Fig. 4

Half of Anthropic's 2025 costs went to R&D, not to serving customers

Anthropic operating expenses by function, 2025, US$ millions

50.1%21.9%11.9%8.9%
  • Research and development$6,338M50.1%
  • Cost of revenue$2,773M21.9%
  • Bartz settlement$1,500M11.9%
  • General and administrative$1,123M8.9%
  • Sales and marketing$911M7.2%
  • Total operating expenses$12,645M100%

From Note 8 of the S-1. R&D is mainly compute for model development, research and internal use, plus research staff. Cost of revenue is substantially all compute for customer-serving workloads. Sales and marketing includes compute for free access to Claude. R&D's share is $6,338M ÷ $12,645M = 50.1%.

Fig. 5

At MiniMax and Z.ai, most compute went to training, not to serving users

AI compute costs by use, training vs. inference, latest disclosed period

Training and researchServing customersMarketing and free use
0%25%50%75%100%MiniMaxUS$ thousands · Jan–Sep 2025Training and research 78.9%21.1%Z.ai (Zhipu)RMB thousands · Jan–Jun 2025Training and research 95.3%

Anthropic does not split its compute by use. Its cost of revenue ($2,773M in 2025), which is substantially all customer-serving compute plus some staff and overhead, was 37.9% of its $7,326M compute bill (derived), so at least about 62% went elsewhere: R&D, internal use and free access.

Fig. 6

Final training runs were a small slice of R&D compute

Estimated share of R&D compute spending on the final training runs of released models: OpenAI in 2024, MiniMax Q4 2024–Q3 2025, Z.ai H2 2024–H1 2025

MiniMax22.6%Z.ai12.3%OpenAI9.6%

Epoch AI estimates. R&D compute is from the IPO filings (MiniMax $141M, Z.ai $216M at 7.2 RMB per US$) and Epoch's earlier estimate for OpenAI ($5B). Final-run costs are Epoch's estimates for models released one quarter after each window. The rest went to experiments, synthetic data and research. Shares are of money spent, not of computing operations.

Fig. 7

Pre-IPO estimates missed $2,945M of Anthropic's 2025 costs

Anthropic 2025: Epoch AI estimate (Feb 2026) vs. S-1 actuals, US$ millions

Epoch AI estimate S-1 actual Gap (S-1 actual − Epoch AI estimate)
0$8B$16BGapTotal operating costs$9,700M$12,645M+$2,945MStaff and other costs$2,900M$5,319M+$2,419MCompute$6,800M$7,326M+$526MRevenue$4,500M$4,585M+$85M

Epoch built its estimate from reporting by The Information before Anthropic filed. S-1 staff and other = personnel $1,839M + Bartz settlement $1,500M + sales, marketing and partnerships $568M + other $1,412M. Epoch's compute estimate is R&D compute ($4,100M) plus inference ($2,700M). The estimate got compute nearly right; most of the miss was in everything else.

Free to republish

Add This Graphic to Your Website

Paste one snippet and every chart, stat and citation above appears on your page. It stays interactive and responsive, and it updates automatically when we revise the data.

  • All sources travel with every card
  • Resizes to fit any column width
  • Light or dark theme, full set or a single card
  • Free with attribution, which is built in
Share this graphic
<div class="graphic-embed" data-graphic="ai-company-costs-breakdown">
  <a href="https://hardware.co/graphics/ai-company-costs-breakdown">AI Company Costs Breakdown: Compute vs. Staff (2025–2026)</a> by hardware.co
</div>
<script async src="https://hardware.co/embed.js"></script>
Auto-resizes to fit its content. Works on WordPress, Webflow, Squarespace, Ghost and plain HTML.
Preview

What this graphic shows

The Compute Ruler lines up four sets of filed accounts from three AI model developers: Anthropic's for 2025 and for the first nine months of 2026, MiniMax's for the first nine months of 2025, and Z.ai's for the first half of 2025. Each strip splits operating costs into compute, staff, marketing and everything else. Compute is drawn first, so its share reads off like a mark on a ruler.

The idea comes from an Epoch AI analysis, popularized by Visual Capitalist, that estimated how three AI companies spend their money. Since then Anthropic has filed to go public, so this graphic replaces the estimates with the companies' own reported numbers wherever a filing exists.

How much do AI companies spend on compute?

Compute took 54.5% to 63.8% of operating costs at the three companies, more than staff, marketing and everything else combined in every period. MiniMax had the highest share, at 63.8% of its $282.672 million in costs for January to September 2025. Anthropic's share was 57.9% in 2025 and 54.5% in the first nine months of 2026. Z.ai's was 57.7% in the first half of 2025.

For these companies, compute mostly means renting chips from cloud providers. Anthropic says it runs its models across AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs. MiniMax and Z.ai buy cloud computing services from third parties, including Alibaba Cloud in MiniMax's case.

How much did Anthropic spend on compute in 2025?

Anthropic spent $7,326 million on compute and infrastructure in 2025, up from $2,522 million in 2024, according to its S-1. That was about 1.6 times its 2025 revenue of $4,585 million. Total operating expenses were $12,645 million, leaving an operating loss of $8,060 million. The company's reported net loss, $41,965 million, was far larger because it includes $34,308 million of non-cash changes in the fair value of convertible notes and similar instruments.

In the nine months to September 30, 2026, compute and infrastructure costs reached $18,808 million on revenue of $33,242 million. Anthropic also disclosed about $900.9 billion of committed future spending, substantially all of it on compute and infrastructure, with about $451.6 billion expected before the end of 2030.

Do AI companies spend more on compute or talent?

They spend far more on compute: staff took 14.1% to 24.0% of operating costs at the three companies, against at least 54.5% for compute. Anthropic's personnel costs were $1,839 million in 2025, including about $1.2 billion of stock-based compensation, compared with $7,326 million for compute. Z.ai had the highest staff share, at 24.0%, and MiniMax spent 17.9%.

Is revenue catching up with AI compute costs?

At Anthropic, yes: compute and infrastructure fell from 251% of revenue in the first quarter of 2025 to 58% in the third quarter of 2026. Compute costs still rose every quarter, from $944 million to $9,863 million, but revenue rose from $376 million to $16,956 million. Revenue first exceeded the compute bill in the first quarter of 2026, at $4,731 million against $3,387 million.

How much of AI compute goes to training models?

Most of it goes to research and development rather than to serving users, but only 9.6% to 22.6% of R&D compute went to the final training runs of released models, according to Epoch AI estimates. At MiniMax, 78.9% of cloud compute spending from January to September 2025 was for training ($142.397 million) rather than inference ($37.988 million). At Z.ai, computing fees in R&D made up 95.3% of all its compute spending in the first half of 2025.

Epoch estimates that only 9.6% of OpenAI's roughly $5 billion of R&D compute in 2024 went to final training runs, against 22.6% at MiniMax and 12.3% at Z.ai. The rest paid for experiments, scaling tests, synthetic data and basic research. At Anthropic, half of 2025 operating expenses ($6,338 million of $12,645 million) were booked as research and development.

Methodology & caveats

Anthropic figures come from its Form S-1/A (Amendment No. 2, dated November 10, 2026), in US dollars. Full-year figures are audited; quarterly and 2026 year-to-date figures are unaudited. A further amendment filed on November 17, 2026 was not checked. MiniMax figures come from its Hong Kong prospectus in US dollars; the nine-month 2025 figures are unaudited. Z.ai figures come from the Hong Kong prospectus of Knowledge Atlas Technology (Zhipu), in Chinese yuan. They are left in yuan, and only shares are compared across companies.

Operating costs mean cost of revenue (or sales) plus operating expenses, including stock-based compensation. Fair-value changes on convertible notes and preferred shares, finance costs, investment income and Z.ai's impairment losses are excluded. Anthropic reports costs by nature, so its compute, personnel and marketing lines are its own. For MiniMax and Z.ai, we summed the compute (cloud or computing service fees), staff (staff or payroll costs plus share-based payments) and marketing line items across their cost tables; Other is the remainder. Categories are not perfectly comparable: Anthropic's compute line includes colocation, leases and power, and Z.ai's R&D depreciation, which may include servers, is counted in Other. Anthropic's 2025 Other includes a one-time $1,500 million settlement of the Bartz copyright case.

Periods differ: Anthropic's full year 2025 and January to September 2026, MiniMax's January to September 2025, and Z.ai's January to June 2025. Percentages are our arithmetic from the filed figures. The final-training-run shares are Epoch AI estimates, and OpenAI's comes from Epoch's earlier estimate of OpenAI's 2024 compute rather than from a filing. Epoch AI's February 2026 estimate of Anthropic's costs, shown in Figure 7, was made before the S-1 and relied on press reporting. Values are nominal and not adjusted for inflation.

Sources

  1. [1]
    Anthropic, PBC (SEC filing). Form S-1/A, Amendment No. 2 (Nov 10, 2026): Summary consolidated financial data, quarterly results and Note 8https://www.sec.gov/Archives/edgar/data/2130357/000162828026072484/anthropicpbc-sx1ax2.htm
  2. [2]
    MiniMax Group Inc. (HKEX prospectus). Global Offering prospectus (Dec 31, 2025): Financial Information and Accountant's Reporthttps://www1.hkexnews.hk/listedco/listconews/sehk/2025/1231/2025123100025.pdf
  3. [3]
    Knowledge Atlas Technology / Z.ai (HKEX prospectus). Global Offering prospectus (Dec 30, 2025): Summary and Financial Informationhttps://www.hkexnews.hk/listedco/listconews/sehk/2025/1230/2025123000017.pdf
  4. [4]
    Epoch AI. Final training runs account for a minority of R&D compute spending (Gradient Updates, Mar 23, 2026)https://epoch.ai/gradient-updates/r-and-d-vs-training-compute
  5. [5]
    Epoch AI. Compute accounts for the majority of expenses of AI companies (Data Insight, Feb 4, 2026)https://epoch.ai/data-insights/company-spending-breakdown
Cite this graphichardware.co. “AI Company Costs Breakdown: Compute vs. Staff (2025–2026).” hardware.co Graphics, updated 2026-11-17. https://hardware.co/graphics/ai-company-costs-breakdown