TwelveLabs rakes in $100m as video AI moves toward enterprise intelligence

Twelve Labs, Inc., operating as TwelveLabs, has raised $100 million in Series B funding to expand its video intelligence models into a broader enterprise artificial intelligence platform. New Enterprise Associates and NAVER Ventures co-led the round, with participation from Amazon.com, Inc. (NASDAQ: AMZN), Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital and Red Bull Ventures. The San Francisco company plans to increase research and development spending, expand its operations in San Francisco and Seoul, and open offices in New York and London. TwelveLabs is also deepening its relationship with Amazon Web Services through a multiyear agreement covering model distribution and optimisation of video inference workloads on AWS Trainium processors. The strategic question is whether TwelveLabs can turn video understanding from a specialised search capability into an infrastructure layer used across media, government, security, advertising, sports and automotive applications.

The Series B follows a $50 million Series A completed in 2024 and a separate $30 million strategic financing backed by Databricks, Snowflake Ventures, SK Telecom, HubSpot Ventures and In-Q-Tel. Together with the company’s earlier $5 million seed round, TwelveLabs has disclosed at least $185 million in financing across these transactions.

TwelveLabs did not disclose its Series B valuation, revenue, customer concentration, gross margin or operating losses. Those omissions mean the strength of the investment case cannot yet be judged using conventional software metrics, despite the strategic quality of the investor group.

Why does TwelveLabs’ $100 million funding round matter beyond another enterprise AI raise?

Most generative artificial intelligence platforms were initially designed around text. They can read documents, answer questions, generate software code and summarise written material because language can be divided into relatively manageable units and processed sequentially.

Video is considerably more complex. A video contains moving images, spoken words, background sounds, on-screen text, physical actions, facial expressions, objects and events that change over time. Understanding one frame may provide little information about what happened several seconds earlier or what will happen next.

This creates a large commercial problem for companies that own thousands or millions of hours of footage. Media groups, sports organisations, governments, retailers, manufacturers and security operators may possess valuable video archives, but much of the information remains difficult to search or analyse without manual review.

TwelveLabs is attempting to convert that footage into structured, machine-readable intelligence. Its technology can identify scenes, objects, speech, motion and contextual relationships, allowing users to search video using ordinary language rather than relying only on manually entered titles and tags.

The funding matters because enterprises are beginning to move from small video artificial intelligence experiments toward production systems. Once a broadcaster, public agency or security organisation integrates video intelligence into operational workflows, reliability, accuracy, access control and computing cost become more important than an impressive demonstration.

TwelveLabs must therefore evolve from a research-led model provider into an infrastructure company capable of supporting continuous enterprise workloads. The $100 million round gives the company more resources to make that transition, but it also raises expectations around commercial adoption and financial discipline.

How do Marengo and Pegasus turn unstructured video archives into searchable business data?

TwelveLabs’ platform is built around two principal model families. Marengo creates representations of video that capture information across visual, audio and language signals, allowing users to search footage using text, images, sounds or combinations of those inputs.

A broadcaster could use this capability to locate every scene involving a particular action, location or spoken topic across thousands of programmes. A sports organisation could search for specific plays or player movements, while an automotive company could analyse road footage for unusual events.

Pegasus performs a different function by converting footage into structured descriptions and analytical outputs. It can identify scenes, entities, temporal segments and contextual relationships that can be used by applications or other artificial intelligence systems.

The combination creates a two-stage commercial proposition. Marengo helps customers locate relevant footage, while Pegasus helps interpret and organise what the footage contains.

This distinction is important because video search alone may become a feature offered by larger cloud and software platforms. TwelveLabs is attempting to build a deeper system that maintains knowledge across video collections and supports reasoning across multiple clips, events and time periods.

The company’s newer architecture is designed to create persistent memory rather than treating every user question as an isolated request. This could allow systems to identify patterns across large archives, compare events and generate insights that become more useful as additional footage is processed.

The challenge will be accuracy. Video contains ambiguity, poor lighting, overlapping speech, rapid motion and incomplete context. An incorrect document summary may be inconvenient. An incorrect conclusion drawn from security, government or automotive footage can have far more serious consequences.

Can TwelveLabs create a scalable business by moving from models into full applications?

TwelveLabs originally positioned itself primarily as an application programming interface and model provider for developers. This allowed software companies and enterprise teams to integrate video search and analysis into their own products.

The company is now moving higher in the technology stack by developing complete applications for creators, operators and decision-makers. Its first application-layer product, Rodeo, represents an attempt to place video intelligence directly in the hands of users who may not have software engineering teams.

Moving into applications can increase revenue per customer. Instead of selling only model access or processing capacity, TwelveLabs could charge for workflow software, collaboration tools, enterprise controls and industry-specific capabilities.

The shift also increases competition. A model provider can support many software partners without directly challenging their products. An application company may begin competing with customers or developers that previously built on its infrastructure.

TwelveLabs will need to manage this channel tension carefully. The strongest platform model may involve providing core video intelligence through application programming interfaces while offering direct applications in areas where no partner has created a sufficiently comprehensive product.

Application revenue may also be easier for investors to value than raw model consumption. Subscription-based enterprise software can produce predictable recurring revenue, while usage-based video processing may fluctuate with customer projects and archive-migration schedules.

However, building applications requires sales teams, customer support, workflow design and integration expertise. TwelveLabs could become more commercially valuable by moving up the stack, but it could also become organisationally more complex and expensive.

Why is the Amazon Web Services alliance central to TwelveLabs’ expansion strategy?

Amazon Web Services is both an investor and an infrastructure partner. TwelveLabs’ models are available through Amazon Bedrock, allowing enterprise customers to access video search, classification, summarisation and insight-extraction capabilities within an existing cloud environment.

This distribution arrangement lowers a major barrier to enterprise adoption. Large companies may be reluctant to establish new security, procurement and billing relationships with a private artificial intelligence startup, even when the technology is compelling.

Availability through Amazon Bedrock allows customers to use TwelveLabs within infrastructure they may already trust and understand. It also gives TwelveLabs access to Amazon Web Services’ global sales organisation, cloud marketplace and enterprise customer base.

The companies are deepening the relationship through a multiyear commitment to optimise TwelveLabs workloads on AWS Trainium processors. Video analysis can require substantial computing because models must process sequences of images alongside sound and language.

Reducing inference cost will be essential to commercial adoption. A media archive containing millions of hours of footage may be technically searchable, but the project will not proceed if processing costs exceed the business value created.

AWS Trainium could help TwelveLabs reduce dependence on a single graphics-processor supplier and improve the economics of large-scale video workloads. Amazon Web Services also gains another specialised model provider that can increase demand for its custom artificial intelligence chips.

The relationship nevertheless creates strategic dependence. Amazon Web Services controls the infrastructure, customer channel and commercial environment through which many TwelveLabs models will be consumed.

TwelveLabs must preserve enough product differentiation and customer ownership to avoid becoming a feature inside Amazon Bedrock. The alliance is most valuable when it accelerates distribution without weakening TwelveLabs’ ability to build its own brand, applications and direct enterprise relationships.

Which industries could generate the strongest demand for TwelveLabs video intelligence?

Media and entertainment provide the most immediate commercial opportunity because broadcasters, studios and streaming companies own large archives that are expensive to classify manually.

Video intelligence can help these companies locate scenes, create highlight packages, identify licensing opportunities and produce metadata for search and recommendation systems. Older archives may contain commercially valuable material that remains effectively invisible because it lacks detailed descriptions.

Advertising represents another potential market. Brands and agencies can analyse content, product appearances, audience context and campaign footage without relying exclusively on manual tagging.

Sports organisations could use TwelveLabs to search historical footage, identify particular plays, create clips and analyse tactical patterns. The value may extend beyond content production into coaching, scouting and performance evaluation.

Government and public-sector applications may produce larger and longer-term contracts. Agencies hold footage from transportation networks, public facilities, defence systems, body cameras and emergency operations.

These deployments can become mission critical, but they also bring stricter procurement, security and responsible-use requirements. TwelveLabs must demonstrate that its systems can operate within controlled environments and that customers retain authority over sensitive data.

Security applications may involve analysing footage for anomalies or locating events after an incident. Automotive customers could use video understanding to examine road environments, test driver-assistance systems and classify unusual driving scenarios.

The broad addressable market is attractive, but TwelveLabs should avoid attempting to build every industry solution internally. A platform strategy supported by specialist partners may scale more effectively than creating separate sales and product organisations for each sector.

Could video intelligence become an infrastructure category rather than a temporary AI feature?

TwelveLabs’ long-term value depends on whether video understanding becomes a distinct infrastructure layer. If video search and summarisation become standard features inside large multimodal models, specialised providers may struggle to defend premium pricing.

The company argues that models trained specifically around video can understand temporal relationships more effectively than language models that inspect selected frames. This technical difference could matter for applications where actions, sequence and timing are essential.

A video-native system may recognise not only that certain objects appeared, but how they interacted over time. That capability could support deeper analysis than a model examining a small sample of images extracted from a clip.

The commercial moat will not come from model performance alone. Artificial intelligence models improve rapidly, and technical advantages can narrow as competitors invest.

TwelveLabs must build defensibility through enterprise integrations, proprietary workflow data, customer trust, distribution partnerships and the cost of moving large indexed archives to another provider.

Persistent video memory could create particularly strong switching costs. Once an organisation has processed, structured and connected years of footage, replacing the intelligence layer may require expensive re-indexing and application changes.

However, customers will resist becoming trapped in another proprietary platform. TwelveLabs must balance defensibility with interoperability, allowing enterprises to connect video intelligence to existing data platforms, cloud systems and artificial intelligence agents.

What risks could prevent TwelveLabs from converting technical leadership into durable revenue?

Computing cost is the first major risk. Processing video is more expensive than processing many text workloads because models must analyse large quantities of visual and audio information across time.

Customers may recognise the value of video intelligence but limit deployment if the cost of indexing archives or analysing live streams remains too high. TwelveLabs must continuously improve efficiency as video volumes increase.

Competition is another risk. Amazon, Google, Microsoft and other large technology groups already operate cloud, media, security and artificial intelligence platforms. They can bundle video capabilities with storage, databases and broader enterprise contracts.

Specialised competitors may also target particular industries with products tailored for media, sports, surveillance or automotive use. TwelveLabs must remain broad enough to support several markets without becoming less useful than focused vertical platforms.

Data governance presents a significant challenge. Video can contain faces, voices, locations, personal behaviour and confidential commercial activity.

Governments and companies will require strong controls over storage, access, model training and data retention. Regulations governing biometric information and automated surveillance could restrict certain applications or increase compliance costs.

Model errors create reputational and liability risk. Systems may misidentify individuals, overlook important events or generate confident conclusions from incomplete footage.

TwelveLabs must provide customers with confidence indicators, audit trails and human-review mechanisms, particularly in high-stakes environments. Selling an artificial intelligence system is relatively easy when it creates clips for entertainment. It is considerably harder when the output may influence a security or government decision.

What do Amazon and NAVER share-price trends suggest about investor sentiment?

Amazon.com shares traded near $244 during the July 1 session, giving the company a market capitalisation of more than $2.6 trillion. The stock had gained approximately 2.5% over the previous week but remained below its May peak.

Amazon shares were trading within a 52-week range of approximately $196 to $278.56. The stock’s recovery from the annual low suggests continued confidence in cloud and artificial intelligence spending, although investors remain sensitive to the enormous capital required to expand data-centre capacity.

TwelveLabs is financially immaterial to Amazon.com at its present scale. The strategic value lies in strengthening Amazon Bedrock and increasing demand for AWS Trainium processors.

NAVER Corporation closed at approximately KRW197,400 on July 1. The shares had declined about 1% over five trading days and nearly 30% over one month, placing the stock close to the lower end of its KRW190,300 to KRW304,000 52-week range.

The weakness reflects wider questions surrounding NAVER Corporation’s growth, artificial intelligence investment requirements and competition across search, content and commerce. NAVER Ventures’ decision to co-lead the TwelveLabs round demonstrates that the group continues to invest in external artificial intelligence platforms despite near-term pressure on its public valuation.

For NAVER Corporation, TwelveLabs may provide strategic exposure to video intelligence that can complement search, content, cloud and digital-media operations. However, a private investment of this size is unlikely to alter public-market sentiment without evidence of meaningful commercial integration or future valuation gains.

What milestones should investors watch after TwelveLabs closes its Series B round?

The first milestone is commercial disclosure. TwelveLabs must eventually provide stronger evidence of revenue growth, customer retention and the contribution from media, government and other industries.

Usage through Amazon Bedrock will be an important indicator. Growth within the platform could validate Amazon Web Services as a distribution channel and demonstrate that customers are moving from trials to production workloads.

The performance of Marengo 3.0 and Pegasus 1.5 will also matter. Technical progress must translate into lower processing costs, better accuracy and faster enterprise deployment.

Rodeo will test whether TwelveLabs can move successfully into application software. Customer adoption would support a higher-value business model, while weak demand could suggest the company is more effective as an infrastructure supplier.

International expansion must produce commercial results rather than additional overhead. New York and London offices should help the company serve media, advertising, government and enterprise customers in major markets.

Hiring in San Francisco and Seoul will provide clues about priorities. Research spending can maintain technical leadership, but go-to-market expansion is necessary to convert models into recurring revenue.

The company’s next financing event will become another valuation test. TwelveLabs now has enough capital to show whether video intelligence can mature into a durable enterprise category rather than remaining a technically impressive feature awaiting a sufficiently profitable use case.

Key takeaways on what TwelveLabs’ $100 million funding means for enterprise video AI

• TwelveLabs has raised $100 million in Series B funding co-led by New Enterprise Associates and NAVER Ventures.
• Amazon.com, Radical Ventures, Korea Investment Partners, Index Ventures, Quadrille Capital and Red Bull Ventures also participated.
• The company has disclosed at least $185 million across its seed, Series A, strategic and Series B financing rounds.
• TwelveLabs plans to expand research in San Francisco and Seoul while opening commercial offices in New York and London.
• Marengo provides multimodal video search, while Pegasus converts footage into structured information for analysis and reasoning.
• The company is moving beyond model access by developing complete applications, beginning with its Rodeo product.
• Amazon Web Services is TwelveLabs’ preferred cloud provider and distributes its models through Amazon Bedrock.
• Optimising video inference on AWS Trainium could lower processing costs but increases TwelveLabs’ strategic reliance on Amazon Web Services.
• Media, government, security, sports, advertising and automotive customers offer large opportunities but create different regulatory and accuracy requirements.
• Long-term value will depend on revenue growth, inference economics, customer retention and whether video intelligence becomes a standalone infrastructure category.