TwelveLabs' Video Foundation Models Land a $167 Million Bet on the Unwatched Hour

The problem with video isn't making it. It's finding anything inside it. For a sports league like Maple Leaf Sports & Entertainment, a 16-hour search for a specific highlight across an archive used to be a full workday. Now, according to a case study, it's nine minutes [SVG Play, 2026]. That 99% reduction in search time is the kind of unit economics that gets a climate and energy editor to look up from his spreadsheets. It’s a direct translation of compute into saved human hours, and it’s the wedge for TwelveLabs, a San Francisco and Seoul-based startup that has raised roughly $167 million to build video-native foundation models [ngram.com, 2026].

Their bet is simple, if technically daunting: the world is drowning in video, and text-based search,relying on transcripts or manual tags,is hopelessly inadequate for finding visual concepts, actions, or objects. TwelveLabs is selling the ability to ask a video library, in plain language, for the first touchdown, the moment a person in a red shirt entered a room, or every scene with a specific machine sound [TwelveLabs, Unknown].

A wedge of video-native understanding

TwelveLabs approaches video as its own modality, not as a series of still images or a transcript with timestamps. The company’s technology is built around two core proprietary models. Marengo handles multimodal video embedding and semantic search, mapping natural language queries to what’s happening on screen. Pegasus is for video-language generation and understanding, powering tasks like summarization, chaptering, and highlight generation [TwelveLabs, Unknown].

This allows the platform to answer queries that would baffle a transcript search. The difference is between searching for the word "touchdown" in a log and asking an AI to "find the first touchdown of the game where the quarterback scrambles left." The latter requires understanding player movement, the game context, and the visual outcome of a play. For enterprise customers sitting on petabytes of untapped video,from media archives and manufacturing floor footage to security camera feeds,this capability turns a cost center into a searchable asset.

The investor consortium betting on video intelligence

TwelveLabs has assembled a capital table that reads like a who’s who of strategic AI and data infrastructure players. The $50 million Series A in April 2024 was led by New Enterprise Associates (NEA) and included NVentures, Nvidia’s venture arm [Tracxn, Unknown]. A $100 million Series B followed, led by NAVER Ventures [Ventureburn, 2026]. The list of other investors is a signal in itself: Databricks Ventures, Snowflake Ventures, SK Telecom, HubSpot Ventures, In-Q-Tel, and Amazon [Crunchbase, Unknown].

This isn’t just venture capital chasing a trend. It’s a coalition of cloud data platforms (Databricks, Snowflake, Amazon), chipmakers (Nvidia), and industry-specific giants (SK Telecom) placing bets that video intelligence will be a core workload on their stacks. The post-Series A valuation was estimated between $200 million and $300 million [Mapping Studio, 2026]. The subsequent, larger Series B suggests that valuation has climbed significantly as the company proves its enterprise fit.

Seed (2022) | 17 | M USD
Series A (2024) | 50 | M USD
Series B (2025) | 100 | M USD

Traction beyond the tech demo

The path to market is through developers and large enterprises. TwelveLabs offers a self-serve Video Intelligence API platform, and its models are also available through cloud ecosystems like Amazon Bedrock, giving it distribution to existing AI/ML customers [AWS, Unknown]. Target industries are sprawling, from media and sports to manufacturing, security, and education [TwelveLabs, Unknown].

The MLSE case study is the clearest public proof point of traction. Reducing a 16-hour manual process to minutes is a compelling ROI story for any content-heavy business. While specific revenue figures aren’t public, the strategic investor mix and the scale of funding suggest TwelveLabs is moving beyond early adopters and into scaled enterprise deals. The company’s headcount growth is also implied by its active hiring for roles like Solutions Architect and Director of Technical Program Management [Ashby, 2026].

Where the competitive pressure builds

The ambition is large, and so is the field of well-funded competitors. The race to understand and generate video is one of the most crowded in AI. TwelveLabs must navigate a landscape that includes:

• Generative video giants. Runway, Pika, and Google’s Veo are focused on video creation, but their underlying models represent massive R&D investments that could expand into analysis.
• Tech platform expansion. Any major cloud provider could decide to build or buy similar video search capabilities as a native service.
• Open-source models. The long-term risk is that capable video understanding models become commoditized, squeezing the value of a proprietary API.

TwelveLabs’ answer to this is its video-native foundation and its focus on the enterprise search-and-analysis workflow, not just generation. The strategic investments from Databricks and Snowflake are particularly telling; they are bets that TwelveLabs will become the preferred video intelligence layer for companies that already store their data on those platforms, creating a potential integration moat.

The next twelve months

The key milestones to watch will be less about model benchmarks and more about commercial footprint. The company needs to convert its impressive investor roster into a roster of flagship, seven-figure enterprise customers across multiple verticals. Another signal will be deeper, product-level integrations with its strategic investors,imagine a native TwelveLabs video search function inside Databricks or Snowflake’s data interface.

Financially, with a fresh $100 million Series B, the pressure for another raise isn’t immediate. The focus will be on scaling the go-to-market team and proving that the cost of querying their API is justified by the operational savings for customers. If the MLSE story can be replicated across dozens of similar organizations, the unit economics will speak for themselves.

Back-of-the-envelope, if a media company with ten editors each wasting two hours a day on manual video search adopts the platform, that’s 5,000 saved hours per year. At a fully loaded cost of $100 per hour, that’s $500,000 in annual productivity recaptured, a figure that makes a five- or six-figure API bill an easy justification. The incumbent TwelveLabs must beat isn’t another AI startup; it’s the entrenched, expensive, and inefficient human labor of watching video in real-time. For now, the math is on their side.

Sources

1. [TwelveLabs, Unknown] Video Intelligence Platform & API | https://www.twelvelabs.io/
2. [TechCrunch, Dec 2024] TwelveLabs is building AI that can analyze and search through videos | https://techcrunch.com/2024/12/12/twelve-labs-is-building-ai-that-can-analyze-and-search-through-videos/
3. [Tracxn, Unknown] Twelve Labs - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/twelve-labs-62b5
4. [AWS, Unknown] AWS Case Study on TwelveLabs | https://aws.amazon.com/
5. [SVG Play, 2026] MLSE case study on video search | https://www.svgplay.com/
6. [ngram.com, 2026] Funding total report | https://ngram.com/
7. [Ventureburn, 2026] Series B funding report | https://ventureburn.com/
8. [Mapping Studio, 2026] Valuation estimate report | https://mappingstudio.com/
9. [Ashby, 2026] TwelveLabs job listings | https://jobs.ashbyhq.com/twelve-labs
10. [Crunchbase, Unknown] Investor list | https://www.crunchbase.com/