WillItSurvive
YC Fall 2025 · United States

Velvet

4-person YC lab selling multimodal training data and evals to AI labs.

Protection under test
Assessed 27 July 2026 · Evidence as of 27 July 2026
Something worth protecting exists and is being tested right now.
On the recordSince 27 July 2026 · Checkable from by Q3 2026 · Every call on the record Checkpoint: Whether Velvet names its first frontier-lab contract
The call

Velvet is a tiny, recently repositioned team chasing a gap that Surge AI and Mercor are already filling faster.

Velvet sells training data to companies building AI models. The company started because Meta blocked a major competitor (Scale AI) from working with rival AI labs like OpenAI and Google, opening a door for newcomers. Velvet has experienced founders from Meta and Adobe, but it is fighting well-funded rivals like Surge AI and Mercor that already have billions in backing and relationships with the big AI labs. The real test is whether Velvet can land at least one big AI lab customer in the next six months; if not, the window it was built for will have closed.

Why
01
AI pressure: Frontier labs building in-house synthetic data pipelines. As labs generate more of their own training data, the market Velvet sells into shrinks before Velvet proves it can sell into it at all. Could AI do the core job? Only partly.
02
Bigger players: Surge AI and Mercor. Both are already selling data and evals to frontier labs at scale, absorbing the demand Scale AI lost after Meta took its stake. Could a giant copy it? Yes, easily.
03
What protects it. Founders have real build credentials, ex-Meta FAIR and ex-Adobe, but no named client or lock-in has been shown yet.
What would change this call
Velvet names a frontier-lab contract (OpenAI, Anthropic, Google, or Meta) within 6 months. Would show Velvet can convert the Scale AI opening into real revenue, not just positioning.
Surge AI or Mercor announces another acquisition in the multimodal data/evals space. Would show consolidation accelerating and the window closing further on small entrants like Velvet.
Velvet changes its stated business model or positioning again in public listings or job posts. Would show continued instability rather than settling into a defensible niche.
Key numbers
  • Team size: 4 people (2026-03-06)
  • Pre-seed raised: ~$500K (n.d.)
  • Mercor raised $350M at $10B valuation (2025-10)
Named risks
  • No named frontier-lab client confirmed as of mid-2026 despite pivoting to serve them. Severity: high
  • Conflicting and unconfirmed funding figures ($500K to $4M to an implausible $503M) suggest no clear capitalization. Severity: medium
  • Rapid sector consolidation (Sepal acquired by Mercor, Feb 2026) could squeeze out small entrants like Velvet. Severity: high
  • Already pivoted business model once in six months, indicating possible further strategic instability. Severity: medium
  • AI-assisted labeling and synthetic data are eroding demand for human-sourced multimodal data collection. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
Scale AIlosing toMeta's 49% stake (June 2025) reportedly caused frontier labs to reduce Scale purchases, opening the niche Velvet entered
Surge AIgaining onCompany-claimed run rate of roughly $1 billion by mid-2026, absorbing demand displaced from Scale AI (vendor/aggregator-claimed)
Mercorgaining onRaised $350M Series C at $10B valuation Oct 2025 and acquired rival Sepal AI in Feb 2026, far outpacing Velvet's scale
SievecontestingUses identical 'the multimodal data lab' positioning and vendor-claims Fortune 100 clients, directly overlapping Velvet's pitch
Questions people ask
Is Velvet profitable in 2026?
No revenue, customer count, or profitability figures have been disclosed. Velvet is a 4-person YC Fall 2025 startup that pivoted from a consumer AI video editor (launched October 2025) to a B2B multimodal data and evals supplier by March 2026, with no named clients confirmed as of mid-2026.
Who owns Velvet the AI data company?
Velvet is privately held and co-founded by Lucas Mantovani (CEO, ex-Meta FAIR) and Lucas Tucker (CTO, ex-Adobe), backed by Y Combinator's Fall 2025 batch. Sacra reports a roughly $500,000 pre-seed round; no priced institutional round or lead investor has been confirmed as of March 2026.
Did Velvet change its business model?
Yes. Velvet launched via YC in October 2025 as a consumer tool for making AI videos, positioned to rival Adobe Premiere. By March 2026, its own job postings described it as a B2B 'multimodal data lab' selling training datasets and evals to frontier AI labs, a full pivot within about six months.
How much money has Velvet raised?
Figures conflict across sources. Sacra cites roughly $500,000 pre-seed as part of Velvet's YC Fall 2025 batch; a third-party aggregator claims $4.0 million; another claims $503 million, which is almost certainly erroneous. No priced institutional round has been independently confirmed as of mid-2026.
Who is Velvet's biggest competitor?
Velvet competes with Sieve, which uses the identical 'multimodal data lab' tagline and vendor-claims Fortune 100 clients, plus far larger rivals Surge AI and Mercor. Mercor raised a $350 million Series C at a $10 billion valuation in October 2025 and acquired competitor Sepal AI in February 2026.
Could AI replace Velvet's data labeling work?
Partially. Industry commentary as of May 2026 says AI-assisted labeling, combining model suggestions with human review, is now the production norm in multimodal annotation, eroding demand for purely human-sourced data. Velvet's original consumer video product was also flagged as vulnerable to commoditization by cheaper foundation-model video generation.
Timeline
  • 2026-07-16Google Gemini API adds Omni Flash video generation/editing model, raising commoditization pressure on video tools
  • 2026-06-27Velvet site rebrands to 'World Model R&D' under velvet.video and velvetlab.ai
  • 2026-04-04Velvet posts Research Scientist and Founding ML Engineer roles for its data/evals pipeline
  • 2026-03-06Velvet job posting confirms pivot to B2B multimodal data lab and team size of 4
  • 2026-02Mercor acquires competitor Sepal AI, signaling consolidation in training-data sector
  • 2025-10-23Velvet launches via YC as consumer AI video editing tool aiming to replace Adobe Premiere
  • 2025-10Mercor raises $350 million Series C at approximately $10 billion valuation
  • 2025-06Meta takes approximately 49% non-voting stake in Scale AI, prompting frontier labs to reduce Scale purchases
Revision history
27 July 2026
First assessed. Protection under test. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources