WillItSurvive
YC Winter 2026 · United States

Shofo

Shofo scrapes and labels web video into training datasets for AI labs.

Losing its protection
Assessed 27 July 2026 · Evidence as of 27 July 2026
The structural protection is eroding.
On the recordSince 27 July 2026 · Checkable from Q3 2026 · Every call on the record Checkpoint: Deadline for landing named anchor frontier lab contracts to prove the model works
The call

Shofo has a working demo and a shrinking team, with no named paying lab customer yet disclosed.

Shofo takes videos from the internet, labels them, and sells them to AI companies that need training data. The problem is that bigger companies like Scale AI and Labelbox can copy what Shofo does, and the AI labs themselves (like Google and Meta) are building their own labeling tools instead of buying from outside vendors. On top of that, new EU rules are making it harder to use scraped data without permission, which cuts into Shofo's main product. The company has lost about a third of its employees and almost two-thirds of its web traffic in just three months, which suggests it is struggling to keep customers and grow. If Shofo can slow down its employee losses and announce a big customer contract soon, it might prove the doubters wrong, but right now the signs point downward.

Why
01
AI pressure: Frontier labs building in-house video scraping and labeling (Google, Meta, OpenAI, Anthropic). The labs Shofo sells to can build their own scraping and labeling pipelines, cutting Shofo out entirely. Could AI do the core job? Yes, likely.
02
Bigger players: Scale AI, Encord, Labelbox, Surge. Established data labeling firms already sell to the same labs and can add a video crawl lane fast. Could a giant copy it? Yes, easily.
03
What protects it. A founding team with prior crawl infrastructure experience from Correkt and an early public sample dataset that drew real downloads.
What would change this call
Shofo names 2-3 anchor frontier-lab contracts by Q3 2026. Would show real demand and defensibility against lab in-sourcing.
Shofo headcount and web traffic trend by Q3 2026. Continued decline from current 4 staff and -63% traffic signals stalling execution.
A major lab (Google, Meta, OpenAI, Anthropic) announces in-house video-curation tooling. Would show the customer in-sourcing threat materializing.
Key numbers
  • Headcount fell 33% over two years, 6 to 4 (2026-06-25)
  • Raised $500,000 via convertible note (2026-01-01)
  • Hugging Face sample dataset drew 25,000+ downloads (2026-05-10)
Named risks
  • Zero named anchor frontier-lab contracts despite pre-commercial outreach as of June 2026. Severity: high
  • Headcount down 33% and web traffic down 63% QoQ signal stalled execution. Severity: high
  • Frontier labs may build in-house scraping/labeling, eliminating need for Shofo's core deliverable. Severity: high
  • Scraped public video content exposed to anti-scraping enforcement and provenance regulation like EU AI Act. Severity: medium
  • Only $500,000 raised with a 4-person team, likely needs new funding within months. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
Scale AIlosing toScale AI took Meta's ~$14-15B stake June 2025; Shofo has zero named customers as of June 2026
EncordcontestingEncord repositions as Scale alternative post-Meta deal, targeting same labeling customers, May 2026
LabelboxcontestingLabelbox listed as multimodal data-curation peer competing for same AI-lab buyers, per 2026 roundup
Wirestock, Luel, Origin LabcontestingLicensed video/image data marketplaces active in same niche, per 2026-06-13 market roundup
Surge AIcontestingPositioned as managed RLHF/alignment data lab competing in adjacent training-data lane, 2026
Questions people ask
Is Shofo profitable in 2026?
No public revenue has been disclosed. Shofo raised only $500,000 via a convertible note dated January 1, 2026, and its own YC page still solicits computer-vision partners, indicating a pre-revenue, pre-commercial stage as of June 2026.
Has Shofo had layoffs?
Third-party tracker Crustdata reported Shofo's headcount fell from a peak of 6 employees in August 2025 to 4 by June 2026, a 33.3% two-year decline and 20% year-over-year drop, with zero open job listings on YC's jobs page.
Who is Shofo's biggest competitor?
Scale AI is the dominant incumbent in AI training-data labeling; Meta took a roughly 49% non-voting stake in Scale AI for over $14-15 billion in June 2025. Encord, Labelbox, Wirestock, and Surge AI also compete in adjacent annotation and licensed-video lanes.
Could AI replace Shofo?
Shofo's own labeling pipeline is already AI-automated, blending YOLO object detection with VLM reasoning annotation as of May 2026. Its customers, frontier AI labs, are the parties best equipped to build equivalent scraping and labeling in-house, per analysis published May 10, 2026.
Who owns Shofo?
Shofo is privately held, backed by Y Combinator's Winter 2026 batch and angel investor Sahin Boydas, per Crustdata data dated June 25, 2026. No valuation has been publicly disclosed, and total disclosed funding is $500,000 via convertible note as of January 1, 2026.
Does Shofo have any named customers?
No named enterprise or AI-lab customers have been publicly confirmed as of June 2026. Shofo's own YC page still invites computer-vision teams to chat, and it published a free 58,000-video Hugging Face sample dataset that drew 25,000+ downloads as of May 10, 2026.
Timeline
  • 2026-06-25Crustdata reports headcount down 33.3%, web traffic down 63.1% QoQ, growth score 16/100
  • 2026-05-10Shofo publishes 58,000-video Hugging Face sample dataset, draws 25,000+ downloads
  • 2026-01-26Shofo publicly launches on YC Launch as Common Crawl for Video
  • 2026-01-01Shofo raises $500,000 via convertible note
  • 2025-08Shofo headcount peaks at 6 employees
  • 2025-06-01Meta takes ~49% non-voting stake in Scale AI for $14-15 billion, reshaping data-labeling market
  • 2025Shofo founded, pivoting from prior product Correkt into video-data infrastructure
  • 2024-10Team headcount at roughly 2 employees under predecessor product Correkt
Revision history
27 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources