Konnex Raises $18.5 Million To Build The Finance Layer For Physical AI, With Mainnet Next
New York, New York, September 24th, 2026, Chainwire
The world is racing to rebuild how money settles. Konnex is making sure machines that move are part of it.
Konnex, the network that lets robots hire one another, license the intelligence they need for a job, and settle in stablecoins, has raised $18.5 million to build the finance layer for physical AI. The company is taking its purpose built layer one from public testnet to mainnet. On mainnet, robots, AI models and the people who build them will contract, verify and settle physical work in stablecoins.
Its marketplace for robotics behavior models, sensor data and compute is already live, and the chain beneath it has been running end to end since January across 30 subnets. The company is betting that the next constraint on autonomous machines is not how well they perform but how their work is verified and paid for.
The timing is not an accident. Two of the fastest moving markets in technology are converging, and they are converging faster than most people realise.
Everything Is Accelerating
The first is money itself. Web3 has stopped being a speculative sideshow and become the settlement layer the financial world is racing to own. Stablecoin supply has passed $300 billion. Stripe’s settlement chain, Tempo, was valued at $5 billion within months of launch. Circle is now a public company worth tens of billions. Mastercard settles in multiple stablecoins across multiple chains. Stripe, Circle and Tether have collectively poured more than a billion dollars into chains built purely to move digital dollars. Whoever owns settlement owns the future of commerce, and the incumbents know it.
The second is physical AI. Robots that once repeated a single scripted motion now take plain language instructions and turn them into action, powered by vision language action models trained on human demonstration. Robotics and physical AI startups had their best funding quarter on record at the start of 2026. Sensors are cheap, compute is abundant, and every quarter the models get better, the hardware gets cheaper and the fleets get bigger.
These two curves are on a collision course. Within a few years, a drone will inspect a bridge at 3 a.m. and be paid before sunrise. A robotic arm in Rotterdam will license a grasping model from a researcher in Seoul for a single shift. A warehouse fleet will hire a mapping model it has never used before, and settle the bill in seconds. None of this needs a human dispatcher, an invoice or a 30 day payment cycle.
That future needs rails. Konnex is laying them now.
Machines Have Money. They Cannot Touch Anything
Software agents have quietly become paying customers. New standards let an AI agent pay for an API call inside a single web request. Money is no longer the hard part for machines.
Reach is. Every one of those rails ends at the edge of a screen. An agent can buy compute or data, but it cannot hire a forklift, dispatch a drone or confirm that a pallet arrived intact. The economy of digital work has infrastructure that the economy of physical work does not.
There is a second reason to build this now, and it is about who controls the commons. Nvidia agreed this year to acquire Hugging Face, home to some 18 million developers, three million models and half a million datasets, including much of the demonstration data behind early robotics behavior models, for roughly $13 billion. Nvidia has committed to keeping the platform open and hardware neutral, and there is no reason to doubt it. But a commitment is a policy, and policies belong to owners. Konnex’s position is that if physical AI is to remain open and fairly priced over the long run, that openness has to be a property of the protocol itself, one that no single company can revoke.
Money That Waits For Proof
The settlement race so far has been built for software. Paying for physical work requires something digital payments never had to solve: proof that the work actually happened.
A robot can be given a wallet. What it cannot yet be given is a contract whose payout waits for independent validators to check hardware rooted sensor evidence that the pallet arrived, the weld was inspected or the field was mapped. General purpose blockchains move tokens, not torque. Existing decentralised AI networks score text. Physical AI needs a network that can score motion.
In the places where autonomous systems are most valuable, which are also the places with the most liability attached, model quality has stopped being what holds deployment back. A warehouse operator, a hospital pharmacy or an airspace regulator does not primarily need a better model. It needs proof of what the machine did, produced by someone other than the machine’s operator.
“The constraint on scaling autonomy was never the hardware or the models,” says Brandon Torres Declet, founder and CEO of Konnex. “It is that nobody outside the operator can independently verify what a machine actually did. Until that changes, insurers, regulators and counterparties relying on data to make critical business decisions will be reluctant to deploy at scale, treating robots as niche rather than a workforce to be scaled. Konnex makes that evidence a foundational layer of autonomous operations instead of an afterthought, and once the proof is there, the payment can follow it automatically.”
Built For Machines That Move
Konnex is that network: an integration fabric for physical AI. It does not make robots. It makes the market and the payment rails beneath them. It is a Substrate based layer one purpose built for what vision language action models, large behavior models and ROS based robots actually produce: streams of video, GPS, lidar, inertial and torque data. Konnex turns those streams into evidence a contract can settle against.
A Universal Task Language gives any robot and any model a common JSON grammar for work, hardware agnostic from day one, so machines from different manufacturers can coordinate without bespoke integration.
Smart contracts lock the commercial terms (escrow, deadlines, penalties and fee splits) before a single motor turns.
Miners compete to execute each task, supplying the machine, the model that drives it, or both.
Validators apply Proof of Physical Work, an independent check of hardware rooted sensor evidence against the task, scoring results on safety, fidelity and efficiency. Validators compete on measurable performance, earning royalties when they are right and building reputation over time, so the quality of verification is itself priced by a market rather than declared by a vendor. The result is an audit grade receipt that insurers and regulators can rely on.
Stablecoins settle the moment proof clears. If the evidence does not clear, penalties apply. The KNX token secures the network through validator staking, pays protocol fees and supports governance.
Together, these pieces let a task run from start to finish without a human dispatcher: an instruction is specified and hashed on chain as a signed request, miners compete to execute it, validators score the results, and a single proof linking the instruction, the policy trace and the sensor data releases the funds.
“Robots got smart before they got paid,” says Torres Declet. “The world is rebuilding settlement for the digital economy right now, at extraordinary speed. We’re making sure the physical economy isn’t left behind. Once a machine’s work can be proven, it can be paid, and once it can be paid, it becomes a workforce.”
A Community Already Building
Konnex isn’t waiting for mainnet to find its builders. Since the testnet went live in January it has connected more than 380,000 wallets, and its builders have contributed more than 1.5 million human feedback submissions to train robot behavior. Thirty subnets are live across drone navigation, robotic arm control and 3D mapping, each exercising the full task flow from request to settlement.
AI models are racing onto the network. More than ten integrations are in place, including OpenVLA, the π0 family and established SLAM systems, with new integrations landing every month. Builders have an SDK covering a command line interface, Python, an HTTP API and robot connectors, and the reference miner and validator nodes are open source on GitHub for operators to run.
The marketplace is live today. A builder can train a behavior model, publish it, and earn a royalty every time a robot streams it, with payment per call and settlement in stablecoins. The catalog spans large behavior models and vision language action models, robot sensor datasets and decentralized compute, and builders who perform well accumulate reputation and can launch a token economy around their own model.
Where It Shows Up First
The early use cases are deliberately unglamorous. The first commercial frontier is critical infrastructure. Bridges, pipelines and power lines are increasingly inspected by autonomous drone fleets, and the owners, insurers and regulators behind them need proof that every inspection happened and every data point is real. Each inspection becomes a signed task verified independently before payment moves.
Konnex is integrating with enterprise drone dock operators running beyond visual line of sight missions, turning each flight into a verified, self settling contract. The same pattern extends to drones coordinating deliveries, pharmacy robots sharing verified fill data and agricultural machines pooling field data.
Everyone Who Feeds The Flywheel Gets Paid
On top of the stack sits a live marketplace for the entire physical AI supply chain. That includes DePIN operators supplying sensor and connectivity networks, compute providers, developers, owners of demonstration and sensor data, researchers refining models, and the API resellers and routers who package intelligence for others. Each can price what they supply in open competition and earn in stablecoins every time a robot uses it.
That creates a flywheel. Every verified task produces attested real world data, which trains better models, which win more tasks. Today that loop belongs to whoever owns the robot. On Konnex, it belongs to everyone who feeds it, which turns every fleet into a supplier to every other fleet rather than a silo.
The Team
Konnex’s team is more than fifteen specialists across robotics, AI and Web3, including engineers, PhDs and founders with prior companies behind them.
Torres Declet has spent more than 20 years and raised more than $125 million at the intersection of drones, autonomy and regulation. He is a four time chief executive with three exits in the drone and robotics sectors. He founded Measure, an enterprise drone company whose services business was sold to Aerodyne. He then sold its software business to NYSE listed AgEagle Aerial Systems, which he went on to lead as CEO. Most recently he was Chair and Chief Executive Officer of Exyn Technologies, repositioning it from a hardware led robotics vendor into a software first autonomy business. He was appointed to the FAA’s Drone Advisory Committee, held a Department of Defense advisory role, and earlier served as counsel on Capitol Hill and in the NYPD’s counterterrorism bureau.
He is joined by Vladimir Smelov, a blockchain and AI systems engineer with thirteen years in automation and multi agent architectures. The company is advised by Lucas Van Oostrum, who cofounded Nova Sky Stories with Kimbal Musk and helped launch the €100 million RoboValley Fund.
Sooner Than Anyone Thinks
Mainnet deployment is in progress ahead of a token launch. The first mainnet phase brings native stablecoin settlement, dual staking for validators and production grade registries, at which point the KNX token takes on its role in consensus security and incentives. A community campaign built on Konnex Points, quests and soulbound token multipliers has been running since January, and the whitepaper at konnex.world/whitepaper lays out the protocol in nine chapters, from architecture and verification through settlement and the staged path to mainnet.
The internet took two decades to build the rails that let software find, price and pay for other software. The rails for machines that move will not take that long. The models are here, the fleets are scaling and the settlement layer is being rebuilt in real time. The only question is whether physical work gets its own rails, or gets bolted onto rails never built for it.
Konnex is building them now. Hired. Verified. Paid.
About Konnex
Konnex is the finance layer for physical AI. Its purpose built layer one lets robots, AI models and the people who build them contract, verify and settle physical work in stablecoins. Proof of Physical Work turns hardware rooted sensor evidence into audit grade receipts that insurers, regulators and counterparties can trust, and a live marketplace lets builders publish and price behavior models, sensor data and compute in open competition. Since launching its public testnet in January 2026, Konnex has connected more than 380,000 wallets across 30 subnets and integrated more than ten robotics models, with mainnet and the KNX token next. Founded by drone and autonomy veteran Brandon Torres Declet, Konnex is backed by $18.5 million in funding and a team of more than fifteen specialists across robotics, AI and Web3. Learn more at konnex.world.