Mecka AI is reported to be closing a new funding round led by Sequoia Capital at a valuation of about $500 million, according to a TechCrunch story published on 11 September 2026. The report runs to 370 words, rests on two unnamed people, gives no round size, and says the terms are not final and could still change. Mecka AI did not respond to the reporter and Sequoia declined to comment. That is the whole of what is new. Everything else in the headline, the rush for robot training data included, was already on the record in June.

We read the TechCrunch report, the Fortune exclusive that announced the $60 million round on 1 June, two BetaKit stories from June, the Mecka AI website, careers page and privacy policy, the iOS App Store listing, both versions of the EgoVerse paper on arXiv, the egoverse.ai dataset page, the partner data explorer that Mecka AI hosts, the GitHub repository, and the Framework Ventures site. Teams that buy or budget for physical AI data will care about the same thing we did: which numbers are stated by the company, which are reported by sources, and which disagree with each other.

Three disagree. The founding year is 2024 in one account and 2025 in two others. The EgoVerse dataset is 1,362 hours in the paper, 4,003 hours on its website and 4,105 hours in the explorer. The staff count is 40 in Fortune and 45 in BetaKit a day later. None of that undermines the round. It is what a buyer should have in hand before the round closes. We covered the human-motion side of this market when a humanoid robot learned to sprint from motion-capture data, and the funding side when Sequoia doubled down on Cymphony.

What TechCrunch Actually Reported on 11 September

mecka ai 500m valuation sequoia robot training data b moka pot stovetop coffee maker

The story is by Marina Temkin, published at 15:58 Pacific time and last modified about an hour later. Its claims split cleanly into three groups: things two sources told the reporter, things the company had already said in June, and things the reporter could not find out. Reading the story with that split in mind changes what the headline means.

The claims, sorted by who is standing behind them

Claim in the reportWho stands behind itStatus
Sequoia is leading a new round“Two people with knowledge of the deal”Reported, unconfirmed
Valuation of about $500 millionSame two peopleReported, “about”, not final
Size of the roundNobody“TechCrunch has not learned the precise size”
$60 million raised, Framework Ventures leadMecka AI, via Fortune, 1 JuneStated by the company
$100 million run rate by end of 2026Josh Gao to Fortune, 1 JuneProjection, “based on already-signed contracts”
Co-founded in 2024 by four entrepreneursTechCrunchDisagrees with Fortune and BetaKit (2025)
Customer listNobody“Hasn’t publicly disclosed”
Comment from either partyNoneMecka AI did not respond; Sequoia declined

What the 370 words leave out

There is no round size, no statement of whether $500 million is pre-money or post-money, no Sequoia partner named, no use of proceeds and no current revenue figure. The only revenue number in the story is the June projection, which the reporter attributes to Fortune rather than to a fresh conversation.

The two companies TechCrunch names as peers, XDOF and Micro1, are both linked to earlier TechCrunch stories, and the third, Scale AI, is linked to a story about its Meta deal. So the “rush for robot training data” in the headline is the reporter’s own framing of three prior reports, which is fair, but it is not new information about Mecka AI.

The Mecka AI Funding Record, Round by Round

mecka ai 500m valuation sequoia robot training data c pair of thick framed eyeglasses standing upright

The $60 million that every story repeats is not one round. Fortune’s 1 June exclusive, by Ben Weiss, describes a $25 million Series A that closed in November 2025 and a $35 million follow-on. BetaKit, reporting the same day from the Canadian side, calls the second piece a Series A extension raised “in recent weeks” and says both were structured as SAFE financings. Neither outlet gives a valuation for either piece, and Mecka AI has never stated one.

The dated record

DateEventAmountSource
Nov 2025Series A closes, Framework Ventures leads$25MFortune; BetaKit
Spring 2026Follow-on or extension, SAFE structure$35MFortune; BetaKit
1 Jun 2026Both pieces announced as $60M; Mecka AI site sitemap last modified the same day$60M totalFortune; mecka.ai
26 Jun 2026Framework Ventures announces a $400M fourth fund, expanding into robotics and AI$400M fundFortune, via framework.ventures
29 Jun 2026Acquisition of Docula, a three-person medical-billing data startup, disclosed as an early-2026 dealUndisclosedBetaKit
11 Sep 2026Sequoia-led round reported at about $500M valuationNot reportedTechCrunch

The arithmetic the record allows

Three sums can be done with stated figures and nothing else. A $500 million valuation is 8.3 times the $60 million Mecka AI has raised in total. It is 5.0 times the $100 million run rate the company projected for the end of 2026, a figure that is a projection from signed contracts rather than reported revenue. And 102 days separate the 1 June announcement from the 11 September report, which is the “three months” in TechCrunch’s text.

One more number comes from the lead investor rather than the company. Framework Ventures describes itself on its own site as a San Francisco firm whose “check sizes typically range from $5m to $50m”. The $60 million total therefore exceeds the firm’s stated maximum single cheque, which is consistent with two tranches and four co-investors, and is a reason not to read the whole $60 million as Framework money. Framework’s page for the Mecka AI announcement is a headline and navigation, 53 words in total, and points to Fortune for the substance.

Who Mecka AI Says It Is, and When It Started

mecka ai 500m valuation sequoia robot training data d miners pickaxe lying across the plinth

Mecka AI describes itself on its homepage as “the data and deployment layer for physical AI” and, in its own FAQ, says plainly: “We don’t build robots, we’re the integrator.” The legal entity behind the brand is Carbon Based Technology Corporation, trading as Mecka AI, according to the privacy policy dated 20 March 2026, and the same corporation is the seller of record on the App Store. The name Mecka, TechCrunch reports, comes from “mecha”, the fictional giant robot controlled by a human.

The founding date, in five sources

SourceWhat it says about the startStaff and location
TechCrunch, 11 Sep 2026“Co-founded in 2024 by four entrepreneurs”; “two-year-old startup”Not stated
Fortune, 1 Jun 2026“In 2025, after collecting thousands upon thousands of gigabytes of data, the four cofounders decided that they had enough traction to start Mecka AI”40 employees; New York City
BetaKit, 2 Jun 2026Formed “last year”, i.e. 202545 staff, 40 of them in Toronto; offices in Toronto and New York
BetaKit, 29 Jun 2026Docula acquired in early 202645 people, majority Canadian; New York office opened June 2026
Mecka AI’s own footprintX account created January 2024; privacy policy effective 20 Mar 2026; iOS app first released 16 Feb 2026Careers page: 17 open roles across the United States, Canada and China

The two dates are reconcilable if you read Fortune literally: data collection began before the company did, which is what “after collecting thousands upon thousands of gigabytes” says. The X account’s January 2024 creation date fits a 2024 start to the project and a 2025 incorporation decision. TechCrunch’s “two-year-old” therefore counts from the project, and Fortune’s 2025 counts from the company. Neither outlet says which, and Mecka AI has not said either.

Four founders, none from robotics

TechCrunch and Fortune agree on the founders and their backgrounds. Josh Gao, the CEO, and Mogen Cheng are Canadians who previously built a restaurant payments startup and sold it. Jason Chong sold a crypto exchange to Coinbase and joined the company. Duy Nguyen, the only non-Canadian, runs operations and, in Fortune’s telling, made his first money reselling sneakers. TechCrunch’s line is that “the four co-founders don’t have backgrounds in robotics”, which the company’s careers page quietly confirms: the robotics expertise is being hired, six roles at a time.

Where the company actually sits

Fortune and the investor’s own announcement call Mecka AI “New York-based”. BetaKit, a day later, counted 40 of 45 staff in Toronto. The New York office opened in June, the month of the announcement. Both are true at once, and the careers page now lists roles in New York, Toronto and Shenzhen, so the geography is three cities, not one.

Mecka AI Sells Egocentric Data. Here Is What That Means

mecka ai 500m valuation sequoia robot training data e tall conifer tree with three tiers of foliage

The product is human demonstrations recorded from the first-person view. TechCrunch’s description is the plainest: Mecka AI “pays people to record themselves performing everyday tasks, like making coffee or fixing cars, using body sensors and smartphones.” Fortune frames the method as the alternative to teleoperation, in which a person drives a robot arm by hand to generate examples. BetaKit adds the scale claim that no other outlet repeats: iPhones and custom cameras shipped to “hundreds of thousands of contributors across 12 countries”, and an internal “video understanding lab”, a computer vision team in all but name, that turns raw footage into training-ready data.

Four ways to get a robot its training data

ApproachWho operatesWhat is capturedWho is selling it
Egocentric human dataA person doing the task with their own handsFirst-person video, head pose, hand motionMecka AI (body sensors and iPhones); XDOF’s wearable tier
TeleoperationA person driving a real robotRobot joint trajectories that match the robot’s own bodyXDOF (GELLO controllers), 130,000 trajectories in its ABC release
SimulationSoftware agents building virtual scenesSynthetic episodes at volume, transfer gap to hardwareResearch labs; see our earlier coverage of agent-built playgrounds
Video-game and controller dataPlayers, unknowinglyMillions of hours of input timing and spatial reasoningGeneral Intuition, which says a model then needs only “a few minutes” of real robot data

The fourth row is the one that matters for the valuation. General Intuition raised $320 million at a $2.3 billion valuation in June on the thesis that real-world hours are close to unnecessary once a model generalises. Mecka AI’s whole business, and XDOF’s, is the opposite bet: that the bottleneck is physical-world examples, and that whoever collects the most of them wins. Both cannot be fully right, and a buyer with a training budget is choosing between them whether or not they name the choice. We looked at the simulation route separately when AI agents built virtual playgrounds to feed robots training data.

The iOS app is a lab tool, not a crowd app

The homepage says the mobile app is “Now available on iOS”. The App Store listing says what it is for. The app called Mecka, version 0.2.2, is “an egocentric data recording application built for internal robotics research labs”, in the Utilities category, 34.7 MB, requiring iOS 15.1, first released on 16 February 2026 and last updated on 18 July with “bug fixes”. It has no ratings.

Its headline feature is “auto-start recording when hands enter the frame”. So the paid contributors that TechCrunch and BetaKit describe are not downloading this; they are using hardware Mecka AI ships to them, which is where the Shenzhen roles on the careers page come in.

Three Different Hour Counts for the Same Dataset

mecka ai 500m valuation sequoia robot training data f padded work glove standing upright fingers spread

The dataset Mecka AI points to from its homepage is EgoVerse, and EgoVerse is where the record disagrees with itself most clearly. The paper, arXiv 2604.07607, was posted on 8 April 2026 and revised on 7 July. Its authors span Georgia Tech, Stanford, UC San Diego, ETH Zurich, MIT, Meta Reality Labs Research, Scale AI and Mecka AI, whose Josh Gao and Jason Chong are listed with the company affiliation. The acknowledgements thank “Meta Reality Labs Research, Mecka AI, and Scale AI for data contribution, engineering support, infrastructure assistance”.

Paper, website, explorer

FigurearXiv paper (v1 Apr, v2 Jul)egoverse.ai “Dataset Snapshot”partners.mecka.ai explorer
Hours of human demonstrations1,3624,0034,105.5
Episodes80k~80k446,511
Tasks1,9651,96528,380
Scenes240240Filter only
Unique demonstrators2,0872,087Not shown
Contributing labsAcademic partners plus industry partners“4 research labs + 3 industry partners”abc, eth, mecka, microagi, rl2, scale, song, trace, wang, yam

The website’s snapshot keeps every figure from the paper except the hours, which have grown from 1,362 to 4,003 while episodes, tasks, scenes and demonstrators stay identical. Hours cannot triple without the episodes changing unless the average episode became three times longer, so the likeliest reading is that the hours were updated and the rest was not. The explorer is a different population again: 446,511 episodes and 28,380 tasks, because it filters across ten lab tags rather than the paper’s release. One takeaway sentence before the chart: quote the paper’s figure to a customer and you undercount what the explorer holds by two-thirds.

Hours of human demonstrations, as a share of the largest published count (4,105.5 h)
arXiv paper, 1,362 h 33%
egoverse.ai snapshot, 4,003 h 97%
Mecka AI partner explorer, 4,105.5 h 100%

The devices, and what the paper found

The paper’s academic partners recorded with Project Aria glasses, Meta’s 75-gram research eyewear, while “industry partners contributed to EgoVerse-I with custom-built rigs for scalability and ease of deployment”. There is also a commodity option: an iPhone on a head strap, recording 1080p video at 30 frames per second through the ultrawide camera, with an accompanying app. That is the shape of the Mecka AI collection kit BetaKit describes, and the paper’s industry stream, EgoVerse-I, is described as “nearly 1,400 hours” and “the largest action-labeled egocentric human dataset”.

The finding buyers should read twice is the paper’s own summary: “policy performance generally improves with increased human data, but that effective scaling depends on alignment between human data and robot learning objectives.” In plain terms, more hours help, but only hours that match the task and the robot. That is the paper Mecka AI co-authored, and it is a caveat on the company’s own pitch that volume is the product.

The repository, counted on 12 September

The GaTech-RL2/EgoVerse repository on GitHub was created on 11 February 2025, carries an MIT licence, and on 12 September 2026 showed 545 stars, 55 forks and 183 open issues, with the most recent commit on its main branch dated 31 August. The homepage links to Mecka AI’s partner explorer and to the paper, and the explorer’s dataset showcase lists tasks such as decorating a cake, sewing fabric, planting lemongrass, disassembling a phone and decanting perfume.

The Rush for Robot Training Data, in Numbers

TechCrunch names three peers. Two of them have their own recent TechCrunch stories with figures in them, and one has a peer of its own, so the set can be laid out with stated numbers only. The revenue column is deliberately labelled by type, because the three figures are not the same kind of number.

The peer set, as each company or its investors stated it

CompanyCapital raisedValuationRevenue figure, and what kindDate and source
Mecka AI$60MAbout $500M, reported, not final$100M run rate projected for end of 2026TechCrunch 11 Sep 2026; Fortune 1 Jun 2026
XDOF$70M Series AAbout $1.2B, in talks, 8VC leadingAnnualised revenue “approaching $50 million”TechCrunch 17 Jun and 4 Sep 2026
Micro1$35M Series A$500M post-money$50M ARR, up from $7M at the start of 2025TechCrunch 12 Sep 2025
Scale AI$14.3B from MetaNot stated in the linked storyNot stated; OpenAI and Google cut ties after the Meta dealTechCrunch 29 Aug 2025
General Intuition$320M$2.3BNot stated; trains on video-game dataTechCrunch 8 Jul 2026

Valuation against the revenue each company chose to state

Dividing each reported valuation by the revenue figure that accompanies it gives three multiples that are not comparable in kind, which is exactly the point. Mecka AI’s $500 million is 5.0 times a projection for the end of the year. Micro1’s $500 million was 10 times a current ARR. XDOF’s $1.2 billion is 24 times an annualised figure that was still “approaching” $50 million. The company with the lowest multiple is the one whose revenue number is a forecast, so the ranking flips if the forecast is discounted at all.

Reported valuation divided by the revenue figure each company stated (bar length is the multiple, out of 24)
Mecka AI, $500M over $100M projected run rate 5.0x
Micro1, $500M over $50M ARR 10.0x
XDOF, $1.2B over $50M annualised 24.0x

Valuation against capital raised

The second sum uses only money in and price out. Mecka AI at $500 million on $60 million raised is 8.3 times. Micro1 at $500 million on $35 million was 14.3 times. XDOF at $1.2 billion on $70 million is 17.1 times. On this measure Mecka AI is the cheapest of the three, and it is also the only one whose lead investor is a firm best known for crypto, which is Fortune’s description of Framework Ventures, not ours.

Reported valuation divided by total capital raised (bar length is the multiple, out of 17.1)
Mecka AI, $500M over $60M 8.3x
Micro1, $500M over $35M 14.3x
XDOF, $1.2B over $70M 17.1x

The rivals appear inside the same explorer

The peer framing hides something the explorer shows. The lab filter on Mecka AI’s partner explorer lists ten tags, and three of them are the names of companies TechCrunch or Fortune present as competitors: “scale”, “microagi” and “abc”. Scale AI has two authors on the EgoVerse paper. MicroAGI is the startup Fortune describes offering New Yorkers a free home clean in exchange for cameras recording the cleaners. ABC is the name of the dataset XDOF is releasing with UC Berkeley, though the explorer does not say whether its “abc” rows are that release.

The tag names are all we have, and we present them as tag names, but they suggest the rush is also a consortium, with Mecka AI hosting the shelf that its rivals’ samples sit on.

What Sequoia Has and Has Not Said About Mecka AI

Sequoia’s contribution to the public record is one word: it “declined” to comment. On 12 September the firm’s site returned a 404 for a Mecka company page, which is what you would expect for a deal that has not closed. Sequoia does not, as a rule, publish numbers even when it does write: its post on the Cymphony round, which we counted last week, ran 736 words and used the dollar sign zero times. If a Sequoia post on Mecka AI appears, expect the thesis and not the price.

The investor that did talk

The lead investor from June has said more, and it is worth separating what it said from what was said about it. Vance Spencer, co-founder of Framework Ventures, told Fortune that Mecka AI is “the fastest-growing revenue company that we’ve ever invested in”. Fortune’s own description of Framework is “crypto-focused”, and 25 days after the Mecka AI story Fortune reported the firm’s $400 million fourth fund “as firm expands into robotics and AI”, which the firm’s site lists next to the Mecka AI story in its news column.

Angel investor Ted Xiao, a former Google DeepMind researcher and a founding member of Jeff Bezos’s Project Prometheus, is the only participant in the June round with a robotics research background.

What “led by Sequoia” would change

If the round closes as reported, Mecka AI’s cap table gains a lead that is not crypto-native, a valuation that is the first the company has ever had on the record, and a data point that TechCrunch’s peer stories will cite in turn. What it would not change is any of the numbers above. The hours, the headcount, the founding year and the run rate are the company’s to clarify, and none of them depends on Sequoia.

The Mecka AI Hiring Page Says More Than the Press

The careers page lists 17 open roles across seven teams in “United States, Canada, China”, and the distribution is a better description of what the company is building than the homepage is. Hardware and Robotics is the largest team, with six openings. Three of the 17 roles are in Shenzhen, none of them in software.

Open roles by team, 12 September 2026

TeamRolesTitles and locations
Hardware and Robotics6Calibration Engineer, Camera and Sensors (Shenzhen, Toronto); Tactile Gloves Program Lead (Shenzhen); Forward Deployed Robotics Engineer (New York); Computer Vision and Robotics Navigation Engineer (New York); Research Scientist, Hand Tracking and Manipulation (New York); Application/Platform Engineer, Web and SDK (Toronto)
Platform Engineering and Infrastructure4Data Annotation Lead (North America); Full-Stack Engineer, ML Tooling (Toronto); Senior Product Engineer and Senior Backend Engineer (New York)
Operations2Data Delivery Lead (Toronto, New York); Strategic Project Lead (New York)
Corporate Admin, Finance and Accounting2Executive Assistant and Office Manager (New York); Head of Legal and Compliance (Toronto)
Research1Research Scientist, Spatial AI and Neural Reconstruction (New York)
Field Engineering and Operations1Operations Manager, Hardware Deployment (Shenzhen)
Marketing1Growth Lead (New York)

Six of 17 is 35 per cent of the hiring in hardware, for a company whose FAQ says it does not build robots. The chart below is that table as shares of the 17 roles.

Share of the 17 open roles, by team
Hardware and Robotics, 6 roles 35%
Platform Engineering and Infrastructure, 4 roles 24%
Operations, 2 roles 12%
Corporate Admin, Finance and Accounting, 2 roles 12%
Research, Field Engineering, Marketing, 1 role each 6% each

What the titles say that the pitch does not

Three titles carry information no press story does. A “Tactile Gloves Program Lead” in Shenzhen means Mecka AI is building a glove, which would add touch data to a product that today is video and motion. An “Operations Manager, Hardware Deployment” in Shenzhen and a “Calibration Engineer, Camera and Sensors” split between Shenzhen and Toronto are the “custom cameras” BetaKit mentioned, being made and calibrated in China.

And a “Forward Deployed Robotics Engineer” in New York is the deployment layer from the homepage tagline, staffed: someone who goes to a customer site and makes a robot work there. A “Head of Legal and Compliance” in Toronto, the fourth role worth noting, arrives at the point where contributor consent across 12 countries becomes a board-level question.

What a Buyer of Robot Data Should Check Before the Round Closes

None of the above is a reason to avoid Mecka AI. All of it is a list of questions that cost nothing to ask now and something to discover later. Teams that budget for physical AI data can use the same discipline they would apply to any data management contract: ask which figure is being quoted, from which document, and on what date.

Five questions with a document behind each one

QuestionWhy it mattersWhere the answer currently sits
Which hour count applies to the data you are buying?The three public counts differ by a factor of threearXiv paper; egoverse.ai; partner explorer
Is the data aligned with your robot and your task?The paper says scaling helps only when it isEgoVerse paper, study section
What did contributors consent to?Recording in homes, kitchens and workshops across 12 countriesPrivacy policy dated 20 Mar 2026; contributor terms not public
Is the $100M run rate current or projected?It is a June projection from signed contractsFortune, 1 Jun 2026
Has the Sequoia round closed?The terms “could still change”TechCrunch, 11 Sep 2026; no filing yet

For readers of the valuation rather than buyers of the data

The number to hold on to is not $500 million. It is 102 days, the gap between the last announced round and the reported one, and it is the same shape as XDOF’s 79 days between emerging from stealth and its own reported talks. Rounds at that cadence are priced on trajectory, and the trajectory Mecka AI has put on the record is a projection. That is normal for the category and worth saying plainly.

Frequently Asked Questions

Is the $500 million Mecka AI valuation confirmed?

No. TechCrunch attributed it to two people with knowledge of the deal, described it as “about $500 million”, and said the terms are not final and could change. Mecka AI did not respond to the reporter and Sequoia declined to comment. No round size has been reported.

How much has Mecka AI raised before this round?

$60 million, made up of a $25 million Series A that closed in November 2025 and a $35 million follow-on raised in the weeks before the 1 June 2026 announcement. Framework Ventures led, with Menlo Ventures, SV Angel, Kindred Ventures and Ted Xiao participating. No valuation was disclosed for either piece.

What does Mecka AI actually sell?

Egocentric human demonstration data: first-person video and motion captured while paid contributors perform everyday tasks with body sensors and iPhones, processed into training-ready datasets for robotics labs and frontier model teams. The company says it does not build robots and calls itself the integrator between hardware, models and commercial partners.

Why do the EgoVerse hour counts differ?

The arXiv paper states 1,362 hours for its release. The egoverse.ai snapshot states 4,003 hours with every other figure unchanged, and the partner explorer reports 4,105.5 hours across a larger population of 446,511 episodes from ten lab tags. The most likely reading is that the website’s hours were updated while its other counts were not, but none of the three pages explains the difference.

Who are the Mecka AI competitors named in the coverage?

TechCrunch names XDOF, Scale AI and Micro1. Fortune adds Wayve and MicroAGI as examples of training on captured human or camera data. Two of those names, Scale and MicroAGI, appear as lab tags in the data explorer Mecka AI hosts, and Scale AI has two authors on the EgoVerse paper.

References