Financials

What AI Labs Have Actually Reported in Revenue: 2024 to 2026 Data Breakdown

A rigorous breakdown of GAAP audited revenue, annualized run-rates (ARR), and official filings from Anthropic, OpenAI, Databricks, Cohere, and Scale AI.

By Swet Parvadiya·October 22, 2026

Separating Audited Accounting from Artificial Intelligence Hype

Headline valuations across artificial intelligence have surged into trillion-dollar territory, but separating headline estimates from verified accounting requires examining actual regulatory filings, investor memorandums, and verified corporate disclosures published between 2024 and late 2026.

In high-growth tech sectors, companies and financial media frequently conflate annualized run-rates (ARR)—taking a single peak month or quarter and multiplying by twelve—with actual GAAP full-year revenue collected in the bank. Below is an evidence-based breakdown of what the leading frontier AI model builders, enterprise data lakehouses, and infrastructure platforms have actually reported.

Anthropic: From $386M to $4.59B in Audited Revenue, and a $65B Run-Rate

Anthropic PBC is incorporated in Delaware as a public benefit corporation. While still privately held, disclosures from its confidential draft S-1 prospectus (submitted to the U.S. Securities and Exchange Commission in June 2026 and reviewed by Reuters and Capital Brief in September 2026) provide the first audited window into frontier lab unit economics:

2024 Audited Revenue: $386 million, against an operating loss of $2.98 billion.
2025 Audited Revenue: $4.59 billion (a 1,088% increase), alongside an operating loss of $8.06 billion. Operating expenses totaled $12.65 billion, led by $7.33 billion in compute and cloud infrastructure. Its reported GAAP net loss of $41.97 billion was heavily impacted by a non-cash mark-to-market charge of roughly $34 billion on convertible financing rounds with Amazon and Google.
2026 Run-Rate (ARR): Anthropic disclosed to investors in July 2026 that its annualized revenue run-rate had climbed to about $65 billion (up from $47 billion in May 2026 and $14 billion in February 2026). Second-quarter 2026 revenue was booked at $11.5 billion, accompanied by its first positive adjusted operating income. Its specialized developer agent, Claude Code, crossed an annualized run-rate of $2.5 billion by February 2026.

OpenAI: $20B+ at Year-End 2025, with a Disputed 2026 Horizon

OpenAI operates under a hybrid capped-profit corporate structure partnered with Microsoft. The company reported dramatic revenue scaling through consumer subscriptions (ChatGPT Plus, Pro, Team, Enterprise) and API token billing:

2023 Reported Run-Rate: Approximately $2.0 billion ARR.
2024 Reported Revenue: Approximately $4.0 billion annualized (roughly $1.0 billion per quarter by Q4 2024).
2025 Reported Revenue: Exceeded $20.0 billion annualized run-rate at year-end 2025, as confirmed in corporate partner updates.
2026 Disputed Horizon: Unlike Anthropic's draft S-1 figures, OpenAI's 2026 metrics carry conflicting press reports. Financial press accounts in late September 2026 estimated OpenAI's run-rate at roughly $70 billion when counting gross partner revenue (including Microsoft Azure OpenAI billing), while direct core platform run-rate estimates stood closer to $50 billion. Because primary disclosures have not reconciled this difference, CorpDigest tracks its verified baseline at the company-stated end-2025 figure.

Databricks: Enterprise Lakehouse Acceleration to >$7B ARR

While foundation model builders bear massive training compute bills, enterprise data lakehouses that organize corporate data for AI have captured massive recurring cash flows:

FY2024 (Ended January 31, 2024): $1.60+ billion full-year revenue, marking more than 50% year-over-year growth.
FY2025 (Ended January 31, 2025): $2.60 billion full-year revenue, maintaining a net retention rate above 140%.
Q2 FY2027 Run-Rate (Quarter ended July 31, 2026): Surpassed a $7.0 billion revenue run-rate, accelerating over 80% year over year. The previous Q4 FY2026 run-rate stood at $5.4 billion.

Enterprise AI Specialists: Scale AI and Cohere

Beyond the primary hyperscaler-backed labs, specialized enterprise model builders and data annotation infrastructure platforms have grown substantially:

Cohere: Grew from approximately $13 million in 2023 annualized revenue to $62 million in 2024 ARR, reaching $240 million ARR by late 2025 according to investor disclosures reported by CNBC.
Scale AI: Expanded from an estimated $290 million in 2022 to $335 million in 2023, reaching approximately $870 million in revenue in 2024 (up 160% year-over-year, per Bloomberg and company disclosures).

Summary: Full-Year Audited vs. Annualized Run-Rates

The table below summarizes verified full-year performance versus annualized run-rates:

Company	2024 Revenue	2025 Revenue	2026 Run-Rate (ARR)	Reporting Basis
Anthropic	$386M	$4.59B	~$65B ARR	Audited draft S-1 (Reuters) + CNBC disclosures
OpenAI	~$4.0B (ARR)	>$20.0B (ARR)	$50B–$70B (Disputed)	Company disclosures; disputed partner gross accounting
Databricks	$1.60B	$2.60B	>$7.0B ARR	Official annual results + Q2 FY2027 run-rate release
Scale AI	$870M	—	—	Press-reported estimates (Bloomberg)
Cohere	$62M (ARR)	$240M (ARR)	—	Investor memorandum disclosures (CNBC)

The Takeaway: Run-Rate vs. Cash Reality

Annualized run-rate metrics allow fast-growing startups to project forward momentum, but financial analysts must bear in mind two structural realities:

Run-rates are not GAAP revenue: A laboratory booking a $65 billion annualized run-rate in July does not have $65 billion of trailing cash collected over the preceding twelve months.
Infrastructure obligations anchor unit economics: Frontier model training and inference require multi-billion-dollar compute commitments with providers like Nvidia and cloud hyperscalers. High top-line growth is a prerequisite to surviving the massive capital intensity required to train next-generation models.