Web search, fetch and curl are all blocked in this environment, so I could not pull live figures. I’m writing from reporting through mid-2026, citing only primary-source URLs I’m confident exist, and giving ranges where the exact number is unverifiable.
Scale AI’s most recent priced valuation is about $29 billion, set in June 2025 when Meta paid roughly $14.3 billion for a 49 percent non-voting stake. That more than doubled the $13.8 billion valuation from the May 2024 Series F. No new primary round has been publicly priced since. Growth is the harder story: revenue was reported near $870 million in 2024 with a $2 billion target for 2025, but the Meta deal cost Scale its largest model-lab customers and triggered a 14 percent layoff, so 2025 revenue is widely reported to have landed well below target.
What is Scale AI worth right now?
The $29 billion figure is a post-money mark from a strategic buyer, not a venture round, so it reflects what Meta paid for talent and a commercial relationship as much as a view of Scale’s standalone business. Secondary trades and employee tender offers since then have reportedly cleared below that mark. The valuation history shows the step change:
| Date | Round | Raised | Valuation |
|---|---|---|---|
| Aug 2019 | Series C | $100M | $1B |
| Dec 2020 | Series D | $155M | $3.5B |
| Apr 2021 | Series E | $325M | $7.3B |
| May 2024 | Series F | $1B | $13.8B |
| Jun 2025 | Meta investment | ~$14.3B | ~$29B |
The Series F, led by Accel with Amazon, Meta, Nvidia, AMD, Cisco and Intel participating, is described in Scale’s own announcement. The Meta transaction is described in Scale’s June 2025 post, and the carrying value appears in Meta’s quarterly filings on SEC EDGAR.
How was the Meta deal structured, and why 49 percent?
Meta took non-voting shares, no board control, and a licence to Scale’s data pipeline. Stopping short of 50 percent and of voting rights kept the deal from being a reportable acquisition under the Hart-Scott-Rodino rules administered by the FTC’s Premerger Notification Office. Part of the proceeds went to the balance sheet and part was distributed to existing shareholders and vested employees. Co-founder Alexandr Wang left to run Meta Superintelligence Labs as Chief AI Officer, and Chief Strategy Officer Jason Droege became CEO. Meta also hired a handful of Scale’s senior researchers, which is why many analysts read the price as an acquihire premium.
How fast is Scale actually growing?
Pre-deal, the trajectory was steep. Reported revenue roughly doubled from 2023 to 2024, and management guided to about $2 billion for 2025 on the strength of frontier-lab contracts for reinforcement learning from human feedback and evaluation data. The deal broke that plan within weeks. Google, reported as Scale’s largest customer at around $150 million of planned 2025 spend, moved work to rivals, and OpenAI and xAI wound down. Public reporting since puts 2025 revenue somewhere between $1 billion and $1.5 billion, depending on how much Meta’s expanded contract backfilled. Scale has not published audited numbers, so treat anything more precise as a leak, not a fact.
What changed inside the company after the deal?
In July 2025 Scale cut about 200 full-time staff and 500 contractors, roughly 14 percent of employees, concentrated in the generative-AI data business. Droege’s memo said the unit had grown faster than demand and had too many layers. Three lines of business carry the company now:
- Meta. The largest customer by a wide margin, buying training and evaluation data for Llama-successor models.
- Government. Scale Donovan and the Thunderforge contract for the Department of Defense’s Chief Digital and AI Office, described at ai.mil, plus sovereign-AI deals in the Gulf.
- Enterprise and evaluation. The GenAI Platform, custom agent deployments, and the public SEAL leaderboards, which keep Scale relevant to labs that no longer buy its data.
Who is taking the share Scale lost?
Surge AI, bootstrapped and reportedly past $1 billion in 2024 revenue, picked up much of the frontier-lab RLHF work. Mercor, which recruits domain experts for model evaluation, raised at a $10 billion valuation in late 2025. Handshake, Turing, Micro1 and Invisible all grew on the same shift away from a Meta-affiliated supplier. The common theme: labs want expert-grade data from vendors with no equity ties to a competitor.
Is an IPO or new round coming?
Not soon. A 49 percent strategic holder makes a conventional IPO awkward and a new venture round nearly pointless, since Meta already capitalised the company. The realistic paths are a tender offer for employee liquidity, a Meta buyout of the remaining 51 percent once antitrust appetite allows, or a sale of the data unit while the enterprise and government business continues. Watch Meta’s 10-Q equity-method line each quarter. If the carrying value is written down, that is the only audited signal of where the $29 billion mark really stands.
Why does a production company care?
Because the demand that funded Scale has moved to multimodal data. Labs now pay for licensed, rights-cleared video with shot-level annotation, and the buyers are increasingly Surge, Mercor and the labs directly rather than Scale. For a studio with a deep archive, that is a licensing line worth pricing per clip-hour, with clear releases and metadata, before an intermediary prices it for you.
