Mecka AI has closed a $60 million Series B round backed by Sequoia, Nvidia, Microsoft and other giants, pivoting from a Crypto points project to a physical AI data infrastructure company with an annualized revenue run rate exceeding $100 million.

A project that once used Crypto-style points to attract users to contribute data has now won the backing of tech giants and investment institutions such as Sequoia Capital, Nvidia and Microsoft, and expects revenue of $300 million by the end of the year.

On October 7, Mecka AI announced the completion of a $60 million Series B round led by Sequoia Capital, with participation from new investors including Nvidia, Microsoft's M12, Qualcomm Ventures and Samsung, and continued support from Framework Ventures, Kindred Ventures and Neo, a startup incubator and VC.

The company also disclosed that its annualized revenue run rate had already exceeded $100 million as of June 2026, and is expected to reach $300 million by the end of the year. It is worth noting that this is the company's disclosed annualized revenue run rate, not full-year revenue already realized.

For the Crypto industry, Mecka is not an entirely unfamiliar AI company. It not only has investors from the crypto space, but its founding team also has a Crypto background. More notably, Mecka launched a points program with crypto-industry overtones in its early days, which at one point sparked speculation among users about a future token issuance.

Today, however, Mecka's website and funding announcements barely emphasize Web3 anymore, instead positioning the company as a physical AI (Physical AI) data infrastructure company.

From Crypto points to a robotics data business, what changes has Mecka gone through?

Mecka's four main co-founders are: Josh Gao (CEO), Jason Chong (CTO), Duy Nguyen (COO), and Mogen Cheng (CPO).

Interestingly, none of the four initially had a professional robotics research background. CEO Josh Gao previously co-founded a restaurant software startup with Mogen Cheng, and sold that business in 2023.

Co-founder and CTO Jason Chong, meanwhile, is a familiar face in the Crypto industry. He previously founded the crypto payments startup Utopia Labs, which was later acquired by Coinbase . According to Fortune magazine, Duy Nguyen, who is in charge of operations, has experience in the sneaker resale business, having made millions of dollars flipping sneakers.

Josh Gao acknowledged in a Fortune interview that none of the team members came from the robotics industry. By reading research papers, visiting robotics labs and talking with researchers, they gradually identified a problem: the robotics industry has increasingly advanced hardware and models, but lacks large-scale, real and sufficiently rich human motion data. The team thus settled on a direction: rather than building robots directly, provide robots with the data they need to learn about the real world.

Regarding the founding date, there is some discrepancy in public reporting. TechCrunch traces Mecka's founding back to 2024, while Fortune says the four founders decided to officially launch Mecka AI in 2025 after accumulating a large amount of data in the early stage. Therefore, 2024 can be understood as its early project exploration phase, and 2025 as the company launch phase described in reports.

In terms of funding, Mecka publicly completed an $8 million seed round in August 2025, led by Neo.

In June 2026, the company announced a combined $60 million in funding ($25 million Series A and $35 million in follow-on investment), led by Framework Ventures, with participation from Menlo Ventures, SV Angel, Kindred Ventures and others.

Four months later, Mecka secured another $60 million Series B round, with its investor lineup further expanding to include Sequoia Capital, Nvidia, Microsoft's M12 and other institutions.

From Framework Ventures to Sequoia Capital, and then to strategic investors such as Nvidia and Microsoft, Mecka's funding lineup is no longer confined to Crypto circles.

AI can write articles and generate code, but getting robots to learn to carry coffee, fold clothes and repair cars still requires a great deal of training.

The reason is simple: language models can obtain massive amounts of text from the internet, but robots need to understand motion, force, contact and spatial relationships in the real world. This information cannot be obtained directly from web pages.

What Mecka does is turn everyday human activities into training data that robots can use.

Its operating process is mainly divided into three steps.

First is data collection. Mecka organizes people to use phones or self-developed multi-sensor devices to record everyday activities such as cooking, organizing items and repairing cars.

Then comes data processing. The company uses technologies such as computer vision and multimodal models to process raw video and sensor data, extracting content such as motion trajectories, hand poses and three-dimensional spatial information.

Finally, data delivery. Mecka provides the processed data to robotics labs, AI model teams and enterprises for model training, evaluation and actual robot deployment.

Mecka's core business model is B2B data and related services.

In August 2025, according to Upstarts, Mecka had publicly disclosed paid contracts with some Fortune 100 companies and listed the robotics company 1X Technologies as an early partner.

Currently, the company says it has provided services to multiple frontier robotics labs and several of the US "Magnificent Seven" tech companies, but has not published a complete customer list or specific contract amounts.

However, Mecka is no longer satisfied with just selling data.

In its latest funding announcement, the company also emphasized its robotics commercial deployment business, including hardware integration, on-site data collection, model post-training, system integration and subsequent operations and maintenance.

This means Mecka hopes to move from being a training data supplier to becoming a service provider for enterprises deploying robots.

The value of this model lies in the fact that Mecka can not only sell one-off data, but may also build long-term relationships with customers through deployment and ongoing operations.

Compared with its current physical AI positioning, Mecka's early Crypto overtones were far more pronounced.

In 2024, Mecka launched a points program inspired by the crypto industry. The early points also covered social tasks, serving a user-growth function as well.

Regarding this experience, according to Upstarts, Mecka CEO Josh Gao once offered an explanation. While this model helped the project gain attention in its early days, it also brought trust issues to the company, especially when working with large enterprises. CTO Jason Chong, for his part, believes that if a company wants to collect millions of hours of video data of uneven quality, a Token mechanism might be useful; but if that scale of collection is not needed, it could also become a distraction.

Subsequently, Mecka pivoted to gamified everyday tasks, letting users contribute data through activities such as making the bed. For enterprise-customized tasks, the company provides dedicated equipment to screened data collectors to obtain higher-quality data.

So, does Mecka still give users points now? This cannot yet be confirmed.

In February 2026, Mecka launched a new iOS app on Apple's App Store, which the company officially described as a first-person data recording tool for its internal robotics research lab, mainly used for data recording, uploading and task management, with no public mention of points, Token or airdrop mechanisms.

The author downloaded the app, but an Access Code is required to enter, so it is temporarily impossible to view the internal tasks and reward rules, and thus impossible to determine whether data providers currently receive points, cash or some other form of compensation.

Looking back at Mecka's development, what deserves the most attention is not whether it will issue a token, but how its data collection model has gradually aligned with enterprise needs.

The early points program solved the cold-start problem: how to a