A new medium is valued in two stages: the machines first, and only later what the machines need to run. Streaming is the clearest case. Spotify spent years signing the major record labels before it could open in the United States. A music app with no licensed music is an empty shelf.

Once every service needed the same songs, old catalogues became some of the most sought-after assets in the business. Anyone with engineers could build a music app. Only a few companies held the rights to what it played, and they set the terms.

Artificial intelligence is still in the first stage. Avi Patel, Kled's founder, put figures on it in an essay on 30 September. Four companies will spend more than $700 billion on AI infrastructure in 2026. The whole industry selling human-made data takes in about $15 billion a year.

His argument is that chips arrive on a purchase order and new training data does not. Epoch AI expects the internet's usable text to run out between 2026 and 2032. The data that comes after it has to be collected from people. That takes their consent, proof that each one is real, and a way to pay them in countries a lab's bank does not reach.

Avi ran a music-licensing marketplace before Kled. He is building the same kind of business a second time, for buyers with far deeper pockets.

Kled is that business. Its app pays more than 600,000 people for the photos, video, voice and tasks they make with their phones. Its records carry public receipts on the DATA Network, where they are 98% of the total. Since 15 October labs can also request data on Kled's marketplace.

Kled, the company, last disclosed a valuation of $280M. KLED, the token tied to it, is worth $12.8M fully diluted on 20 October. So the market prices the token at less than a twentieth of the company. Is that gap a fair discount, or is the market pricing the supplier of AI's scarcest input as if it were one more consumer app?

• Kled pays people for the data AI labs are short of. More than 600,000 contributors make photos, video, voice and task data with their phones. Uploads have passed 1.5B, with 5M to 8M more a day.
• Labs can now request that data directly. Kled Studio lets a lab write a task, choose who does it and test the result on its own model. Kled announced it on 1 October, with early access for invited labs. The marketplace at kled.ai/datasets opened on 15 October, and every request needs Kled's approval.
• Demand is ahead of the workforce. On 1 October Avi said the labs want about 400 tasks live at once, against the 200 the app fills comfortably.
• Kled will train its own models. Kled Labs, announced on 12 October, will train models on the contributors' data alone. The goal is to beat closed models at important human tasks.
• Kled's records carry public receipts. On 20 October Kled held 277.6M of the DATA Network's 282.6M receipts, or 98%. It also had 71% of the contributors.
• The network Kled supplies is priced far above KLED. DATA, the network's token, is worth $151.5M fully diluted on 20 October, against $12.8M for KLED. The Sui Foundation announced a $30M investment in the network on 12 October.
• Kled, the company, is valued at $280M. That is the latest valuation Kled has disclosed, in its tokenomics of April 2026. It is 22 times the token's fully diluted value.
• Every KLED trade adds to a stake in the company. All of KLED's trading fees go to a holding company that Street, an independent third party, set up for the token side. It holds a 5% equity position in Kled, $14M at that value. Up to 10% is set aside.
• The app buys and burns KLED. Each dollar a worker converts to XP, the app's level points, buys a dollar of KLED and burns half of it. Kled has also bought back 43.1M KLED from revenue, in March and June.
• Analysts value KLED well above its price. Blockworks Research's Nick Carpinito puts it at $46.5M to $109M. His $66M base case is 5.2 times KLED's value on 20 October.

Kled is an app that pays people for the data they already make with their phones. A contributor picks a paid task, such as a photo of a receipt, a video of a kitchen, a voice recording or a document, and uploads it. Once the balance reaches $25, Kled pays out in cash to PayPal, Venmo or a Solana wallet. Kled bundles the uploads into datasets and licenses them to AI labs, robotics companies and research groups.

Reuters pays reporters for stories and sells each story to many newspapers. Kled works the same way, except that the reporters are ordinary phone owners and the newspapers are AI labs. A licence can be non-exclusive, as Kled's largest announced one is, so the same file can sell to more than one lab. Each record Kled registers also carries a public receipt of who made it, on what terms and what they were paid.

Kled is both the app and the company that runs it, legally Nitrility Inc. KLED is the token Avi launched in May 2025. They are parts of one project. The fees on KLED trades tie the token to the company. Two things tie it to the app: the XP workers buy there and the tokens Kled buys back from its revenue.

Avi founded Kled. He left computer science at the University of Illinois after two weeks in 2021 and ran a music-licensing marketplace. He turned to training data when the AI companies he sold music to kept asking for it. He has said more than 200 venture firms turned him down, and he is still building Kled in public, from New York.

KLED's supply is capped at 1B tokens, and the roughly 999.8M left after burns are all already trading. So its fully diluted value, the price times every token in existence, is also its market value: $12.8M on 20 October. Avi published the token's rules, its tokenomics, on 23 April, and they start from a problem every application token has.

A chain's own coin, such as ETH, is spent as gas and partly burned, so its value tracks use. A token like KLED has no such loop, and a growing app does nothing for it by default.

Kled, Avi wrote, "Kled is a real product valued at $280,000,000 with real users and an end goal of an IPO." So its token has to gain value as the network grows and make the business stronger at the same time. His answer ties KLED to the company's shares on one side and to the app on the other.

On the shares side, Kled has set aside up to 10% of the company for the Street Global Foundation. Of that, 5% is already committed to the token side.

Kled's investors hold equity in Nitrility Inc., the company behind the app. The latest valuation Kled has disclosed, in its tokenomics of April 2026, is $280M. We use that figure throughout.

The token side has equity too. Street set up a holding company to be the token side's shareholder in Kled. We call it Street's holding company. It holds a 5% equity position in Kled, $14M at that valuation. It can take up to another 5% in future funding rounds. Kled sends all of KLED's trading fees to it, and it reinvests them in Kled, building that position toward the full 10%.

When a company is sold or goes public, its token holders are usually cut out. Buyers want a short, clean list of shareholders and one party to deal with. Street built its framework, ERC-S, to keep them in, with three layers. First, Kled owns the business and aims for an IPO. Second, Street's holding company is the one shareholder for the token side, with 5% now and up to 10%. It is paid like any other shareholder if Kled is sold or goes public.

Third, KLED holders sit behind the holding company. After a sale or an IPO they can put proposals to the Street DAO, Street's own voting body, on how the holding company's proceeds are used. The DAO votes on them under its own rules, and Kled has no say.

KLED is the token community's way into Kled's equity. Street's holding company holds that equity for the token side, and every trade, every XP purchase and every buyback adds to it. In June Avi closed the company's shares to new investors until its next formal round, a Series A, and pointed them to KLED.

KLED's other tie is to the app, and it is the one a contributor feels. Kled's tokenomics, Avi's later posts and one holder's proposal describe six uses. Each has its own effect on who holds the token and how much of it exists.

Levels and XP. Workers can convert their cash earnings into XP, which raises their level and opens better-paid tasks. Lab tasks now pay $1,000+ each. For KLED, every dollar converted buys a dollar of KLED on the open market, and half of it is burned for good. So the same purchase adds demand and shrinks the supply.

UFDP. The unburned half pays for UFDP, Kled's free captcha for websites. At ufdp.ai, the usual traffic-light puzzle becomes a real, anonymised data point from Kled's pipeline, shown to at least 50 people at once.

Those who agree with the majority earn KLED. Those who disagree take a hit to their trust score, and their pay multiplier drops. A contributor earns KLED in the browser, and Kled gets a second check on its data. The bought tokens pass to the people doing the checking.

Buybacks from revenue. Kled also buys KLED back from its revenue when it chooses. Blockworks Research's Nick Carpinito traced 43.1M KLED bought in all, 4.3% of the supply, with none sold. That was 8.02M on 31 March and 35.1M over four days from 26 June. The bought tokens fund vested team and contributor grants and UFDP rewards, so they reach people who work for the network over time. Kled says it never sells them.

Payouts and task access. On 1 October Avi set out a plan to make KLED the app's reward currency. That means payouts in KLED, tasks that take KLED to unlock, and payments where dollars are hard to get. For a worker in a country that dollar payouts barely reach, that is a way to be paid. For KLED it would turn the app's 600,000 earners into holders.

Holding to earn more. @asparagoid, who holds KLED, has proposed paying workers slightly more when they hold a minimum of it. It is the way some exchanges cut fees for holders of their token. Kled has not switched it on. It would give every worker a reason to keep some of what they earn.

Compute and storage. Kled's tokenomics name these as next. Kled is exploring letting contributors lend compute to its fraud-detection network. It also plans a suite of tools paid for in KLED, among them compute and data storage.

Kled's stated target is millions of high-level workers converting millions of dollars a month into levels. Against KLED's $12.8M on 20 October, even $1M a month would buy 7.8% of the token's value and burn half of that.

The loop is small today. About 2,000 KLED had burned through XP from April to early October. That is about $26 at today's price, so the loop's size is the number to watch. It grows only as better-paid lab tasks give workers a reason to climb levels. So the labs' demand for Kled's data is also the demand behind the burn.

Kled the company is a startup of about a dozen people. It had raised $14M in all by June and last disclosed a valuation of $280M. It sells licences to its datasets, and it fills buyers' orders for new data through tasks in the app. Avi told Business Insider in March that most of its customers are AI and robotics companies.

Its money comes from contracts with those buyers. In June Kled announced a $12M two-year licence that made its marketplace profitable. On 1 October Avi said contracts now run from $300,000 to $100M depending on the task. Kled publishes no revenue figure, so analysts estimate its revenue from the contracts it announces.

The contracts draw on a workforce that has tripled since March. The app had 200,000 users then, as Business Insider reported, and 500,000 in July. It had almost 600,000 by 1 October, when Avi put the previous day's downloads at about 10,000. By 12 October it had 600,000+. Uploads passed 1B in April and 1.5B since. They come at 5M to 8M a day.

Avi also said on 1 October that Figure, the humanoid robot maker, modelled its own tasks app on Kled's.

V3 of the app, released on 30 September, rebuilt it around the labs. A contributor now sees a ranked feed of paid tasks built from 108 templates, the labs' tasks first. The app takes any kind of data: voice, video, image, text and annotation. A fraud model, FD 1.0, checks each upload before it counts and removes 98.4% of bad actors within their first ten uploads.

Kled gave its reason in the launch thread: "In working with several frontier labs, we found that our existing infrastructure couldn't support many of the complex tasks and modalities needed to advance the frontier."

The next day Kled announced Studio, a desktop app with early access for invited labs, where a lab places its own orders. In Studio a lab will browse Kled's data and write its own task. It will pick who should do it by demographics and test how the data would change its own model before it pays. A custom task can be live in the app within a day, in front of 600,000 people.

Kled has sold data off the shelf since 2025. Studio is made to order: the lab writes the task itself, picks who does it and tests the result before it pays. The contributor who makes each file is still paid in cash.

The limit now is the workforce. On the MCG livestream on 1 October Avi said the labs want about 400 tasks live at once, against the 200 the app's users fill comfortably today. So lab access stays by invitation while Kled works through a smaller group of high-quality requests. "The only rush I have," he said, "is get as many contracts in the app as possible, give users as many options as possible to get paid at a high level."

Kled's tokenomics of 23 April put its data purchase pipeline, the purchases it expects from buyers, above $35M. In June it signed the $12M two-year licence. It also got an $8M request for anonymised financial documents, such as tax returns and credit card reports. Its users can upload those for $50 to $150 a form.

The marketplace opened on 15 October at kled.ai/datasets. There labs browse and request datasets from the contributor network and from a curated set of enterprise partners. Avi was explicit about approval: "all dataset acquisition and listing requests are subject to approval by the kled team. we do not intend to make this process permissionless."

So Kled chooses which buyers and sellers it deals with, the way Coinbase vets every token it lists and every account it opens.

Kled Labs, announced on 12 October, puts that supply to work for Kled itself. Kled will train its own models on its contributors' data alone, with one goal: "to beat closed models at important human tasks." It is hiring a research scientist for benchmarks and evaluations.

Avi says Kled could have started more than six months ago, but chose to wait until its app was ready. He also says "several of the top models will lose to us in important human categories, purely because of our data edge." If Kled's own models do beat closed ones there, the labs it sells to will have the clearest proof of what its data is worth.

Avi launched KLED on the Believe launchpad on 22 May 2025 and has shipped in public since. The iOS app came in September 2025, with about $700,000 paid out in its first five payout batches. In October came Kled ROS, for household robotics footage, and a licence with Maryland Public Television. December brought V2 and a round valuing the company at $100M, and Kled committed part of its shares to Street.

By early January 2026 the app was closing in on 200,000 downloads, and on 15 January KLED reached its high. March brought a $5.5M seed round and the first buyback, and April the tokenomics and 1B uploads. In June Kled signed its largest announced licence, bought back 35.1M KLED over four days and became the flagship of the new DATA Network. By July there were 500,000 people in the app, and V3 shipped at the end of September.

Sixteen months is a long time in this market. From 1 January to 29 September, our bot logged 2,749 tokens making sharp runs. Of those it labelled, 70% rose and faded within the same day. In April Believe's co-founder was arrested. Avi has stayed throughout. He fought that co-founder over the co-founder's sales of KLED, bought tokens back, and kept the token when critics told him to drop it.

A dataset is worth licensing only if the buyer can prove where it came from. A lab that trains on data it had no right to use can be sued for it. Kled's answer is the receipt, and its receipts live on the DATA Network. The DATA Foundation was formerly Story Protocol, a project backed by a16z crypto that had raised $140M. On 25 June it renamed itself, turned its chain into a registry for AI training data and made Kled its flagship.

Kled registers its records on Trace, the network's public registry. Each carries its origin, the licence it was given under and what the contributor was paid. A receipt works like a title deed for a photo. A lab that buys the photo can show a court or a regulator who made it and on what terms.

The labs use Trace. On 1 July Avi wrote that "four multi billion dollar frontier AI labs have already begun using Trace to verify the integrity and provenance of Kled's data before making purchases."

We read Trace's live counts on 20 October. Kled's records numbered 277.6M of the network's 282.6M receipts, or 98%. The next app, Numo, had 5.0M. The count rose by four to seven a second across our reads, in line with Trace's note that every contribution Kled accepts appears within seconds.

Kled also accounts for about 381,900 of the network's 537,200 contributors, or 71%. Nearly every record a lab can check on the registry is Kled's.

Kled has 1.5B+ uploads against 277.6M receipts. So it has yet to register most of what it has collected, and the count has room to grow. It is also the one measure of Kled's output that someone else publishes: the DATA Foundation. It put $3M into Kled in June and Avi is its part-time advisory chief data officer.

The network is like a land registry, and Kled is the developer behind almost every house in it. A registry is worth what is recorded in it, and the developer's houses sell more easily because their deeds are public.

Blockworks' transparency filing on the network names Kled its flagship app and flags its reliance on Kled as a risk. Blockworks Research's study of the onchain AI data stack, released with the DATA Foundation on 30 July, also found Trace's volume concentrated in Kled. From Kled's side, that risk is bargaining power. The network's value rests on Kled's records, and Kled's sales rest on the network's receipts.

On 11 and 12 October the Sui Foundation, which backs the Sui blockchain, announced a $30M strategic investment in the DATA Network and became its largest institutional holder.

The network will retire its own chain and rebuild on Sui. Walrus, Sui's storage network, will keep an encrypted copy of each record that only licensed buyers can open. In April Grayscale had already put Poseidon, the network's data layer, on its list of assets under consideration.

The market values DATA at $151.5M fully diluted on 20 October, down from $213.9M on 12 October. It values KLED at $12.8M. So the registry's token is priced at about 12 times the token of the business that fills it.

The labs buy data from suppliers of every size, and investors value those suppliers in billions. In June 2025 Meta paid $14.3B for 49% of Scale AI, which valued it above $29B. Surge AI made over $1B of revenue in 2024 and was in talks at $25B.

Mercor raised at $10B in October 2025 and has been in talks at $20B. In September micro1 raised at $4B. Snorkel AI tripled its valuation to $3.5B, on $375M a year of revenue at its current pace. Turing was valued at $2.2B in March 2025.

The owners of text are paid too. News Corp gets up to $50M a year from Meta. Disney put $1B into OpenAI with a licence attached. Reddit's non-advertising revenue, led by licensing, went from $15M to $140M in two years.

Avi's case against the largest suppliers is that they process data rather than collect it. They use contractors at laptops, hired from résumés, adding opinions to data that already exists. A phone, in his words, is "there when things happen."

Investors in robot data are backing the same idea with large cheques, because robot makers need video of people doing things with their hands. In October Mecka AI took $60M, led by Sequoia with Nvidia, to pay people to record everyday tasks while wearing body sensors. XDOF raised $70M in June for data from people steering robots remotely.

Skild AI raised about $1.4B at over $14B in January. Physical Intelligence raised $600M at $5.6B in November 2025. Figure, valued at $39B, trains on first-person video from more than 100,000 homes. micro1's chief executive puts robotics companies' spending on such data above $100M a year.

Kled sits in that market with 600,000 people already on its app, a robotics product in Kled ROS, and receipts on its data. Its nearest consumer rival, Luel, raised $31.2M in May with 500,000 contributors. That single round is worth 2.4 times KLED's whole fully diluted value today.

Between 5 and 8 October, Blockworks Research's Nick Carpinito published "Kled Is Grass for Consented Data, at 1/33rd the Size". Grass sells training data to the same labs. His valuation table puts KLED at $46.5M to $109M.

On 9 October Grayscale's Will Ogden Moore called Kled extremely overlooked, citing 1.5B+ uploads and millions in revenue. Like Carpinito, he wrote on a personal account and in his own view. On 12 October he listed KLED among eight tokens, with NEAR, TAO and GRASS. He replied to his own post: "Conviction, long term."

Carpinito compares KLED with Grass, which has about $29M of audited revenue against a market value near $450M, or 15.5 times revenue. He gives KLED half that multiple, 7.75 times, because its holders' claim on the company is weaker. From Kled's public deals he estimates its revenue at $6M to $14M a year.

The $12M licence alone, about $6M a year, puts KLED at $46.5M. His base case, $8.5M a year after cutting 15% for caution, gives $66M. Both public deals delivered give $109M. Against KLED's $12.8M on 20 October, those three cases are 3.6, 5.2 and 8.5 times.

The method has limits. Kled's revenue has not been audited, and the multiple comes from a single peer. In Carpinito's words a revenue audit and a straight answer on what KLED owns would close the gap. His lowest case counts that one licence alone, and it still lands at 3.6 times today's value.

The gap between the company and the token has widened since December. Then a round valued Kled at $100M and the token was worth $25.5M. Kled's latest disclosed valuation, reported in April, is $280M. The token is worth $12.8M on 20 October.

Street's holding company holds a 5% equity position in Kled, $14M at that valuation, and every KLED trade adds to it. So on 20 October the market valued every KLED in existence, $12.8M, below the stake's $14M paper value.

That stake belongs to the holding company, and that is the discount Carpinito prices in. A price below the stake gives no value to the rest of what ties KLED to Kled. That means the XP burn, the buybacks from revenue and Avi's plan to pay the app's 600,000 workers in KLED.

The data suppliers named above are valued at about 8 to over 100 times Kled's own valuation. The registry Kled fills is priced at about 12 times KLED. Say the labs keep paying for consented data with receipts, and Kled keeps routing its fees, its XP and its buybacks through KLED. Then the token has more room to close the gap than the company does.

Streaming taught the music business that whoever holds the licensed supply sets the terms, however many music apps get built. AI is learning the same lesson about data. The labs can buy chips by the truckload. But recordings of people living their lives, given with consent and backed by a receipt, have to be collected one phone at a time.

We think KLED's price treats it as a small app with a weak claim, when its project supplies the input the labs are running short of. We also think the distance between the token and the company is the opportunity.

• The first labs to post tasks in Studio, and the first datasets sold through the marketplace
• KLED as the app's currency: payouts or bonuses in KLED would make 600,000 workers buyers
• Kled Labs' first model, and how it does against closed models on the human tasks Kled picks
• The DATA Network's move to Sui, and how Kled's receipts are valued there

Over the year KLED has slid a long way. On 13 December, when Kled committed part of its shares to Street, it was worth $25.5M. It reached its high on 15 January: $0.0562 during the day, about $56M fully diluted. It closed that day at $41.6M. On 23 April, the day the tokenomics went out, it was worth $14.9M, up from $13.1M three days before.

On 26 June, the day of the $12M licence, it was worth $16.9M, up from $12.3M the day before. Four days of buybacks ended on 29 June with it at $21.9M. V3 shipped on 30 September, and it closed the day at $16.7M. Kled Labs and Sui's $30M came on 12 October, and it closed at $16.1M. The marketplace opened on 15 October, and it closed at $13.3M. On 20 October it was worth $12.8M.