$1.45 Billion for a Robot Model With No Revenue — 23 Times Interlink's Entire Market Cap

Eka Robotics closed a $150 million Series A at a $1.45 billion post-money valuation on Oct 20, 2026, co-led by Andreessen Horowitz and F-Prime — roughly a 10% stake for the new money. The Cambridge, Massachusetts startup, founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, is pre-revenue in any disclosed sense: customers are unnamed industrial pilots running pick-and-place and wire manipulation. The round moves the profit pool out of the arm and into the model, and every other layer of the stack now has to answer for itself.

What makes Eka different from the usual physical-AI pitch is the control signal. Its Vision-Force-Action model deliberately refuses the vision-language-action template that dominates the field. Vision is used for coarse localisation; the grasp itself is driven through contact forces read by a custom, sensor-dense hand. The company exited stealth in April 2026 on that argument, then raised at $1.45 billion six months later.

The argument about who captures the value in embodied AI is usually run on market size. Run it on gross margin instead and it resolves faster. Compute earns software economics. Everything that physically touches the world does not. Across the trailing twelve months, no company that actually builds robot hardware earns a software gross margin.

The ladder is steep enough to matter for the Eka thesis. The model layer is being marked as though it belongs above NVIDIA's compute line. The machine layer is being marked as though it belongs beside a system integrator: Symbotic ships robots into Walmart's network, carries $2.65 billion of trailing revenue, and still converts that into a 21.6% gross margin and $32.9 million of trailing operating income. Deploying the machine is the least profitable activity in the stack, and Eka's valuation says that is the layer being squeezed.

Two comparisons do most of the work. First, Interlink Electronics — the one listed pure-play in force and touch sensing, whose 2025 Form 10-K describes force-sensing resistor and piezoelectric sensor lines serving automotive cockpit, consumer and rugged computing — carries a market capitalisation near $63.5 million on $12.66 million of trailing revenue. Eka's mark is roughly 23 times that entire market value, with no disclosed revenue of its own.

Read down the multiples and a pattern emerges: the more a company touches the physical world, the less the market pays per dollar of revenue. Novanta, which makes force/torque sensors and end-of-arm tooling through the ATI Industrial Automation business it bought in 2021, trades at 4.4 times sales. Interlink, selling tactile sensing outright, trades at 5.0 times. Eka, selling neither, is marked at an uncalculable multiple. The multiple is not a function of revenue — it is a bet on which layer keeps the revenue.

The most revealing detail is that Eka shares its number with SiMa.ai. On Sep 28, 2026, SiMa.ai closed a $150 million Series C at exactly $1.45 billion — the same round size, the same post-money mark — co-led by Fidelity Management & Research and Amplify, lifting that edge-inference chip developer to $1.45 billion from $960 million fourteen months earlier. One company sells a model trained in simulation; the other sells silicon that runs inference on a robot. The market has stopped pricing the layer and started pricing the category.

That flood is the mechanism that turned $1.45 billion into a price rather than a data point. Physical AI took $47.4 billion across 521 deals in the first half of 2026, against $26.4 billion in the same half of 2025 — though nearly a third of it is a single $16 billion Waymo round. Alongside it, Andreessen Horowitz closed a $1.1 billion Machine Age fund in August 2026 explicitly to back hardware startups contributing to physical AI. Two large funds now need model-layer names to price against each other, and a common mark is the cheapest way to clear them.

Here is the part the funding cycle missed. If force becomes the primary control signal, then tactile sensing stops being a safety accessory and becomes the control loop itself. Eka's own hand is the proof: more force sensors than a conventional gripper, not fewer, and priced under $1,000. Hardware deflation and sensor content growth are happening in the same component at the same time, which breaks the usual assumption that a model-layer winner simply commoditifies its suppliers.

The downstream side takes the opposite side of the trade. Cognex is the franchise with the most to lose from a force-first architecture: it posted the best margin expansion in the stack this year, reaching 71% gross margin on $291.3 million of Q2 2026 revenue, up 17%. But its content sits on the vision side of the split Eka is trying to obsolete. Symbotic sits at the other end — holding deployed automation economics at a 21.6% gross margin is the least defensible position in the chain if the intelligence layer prices itself independently of the cell.

The bear case on Eka is not technical. It is that a $1.45 billion mark on a company with unnamed customers behaves like every other physical-AI mark before it: the first credible deployment disclosure either validates the category or resets the comparison. The bull case is that force-first control is a genuinely different primitive, and that the simulator-to-deployment loop is harder to copy than a vision-language-action checkpoint.

The confirmed facts are these: a $150 million round closed on Oct 20, 2026 at a $1.45 billion post-money mark; a $1,000 gripper is claimed to move 1,500 objects an hour at better than 95% reliability; and the listed stack beneath it runs gross margins from 74.7% at compute down to 21.6% at deployed automation, with the robot segment of the largest listed automation vendor losing money outright.

The inference is narrower than the headlines suggest. This is not a verdict that robots are commodities tomorrow. It is a verdict on where the next dollar of margin attaches. Eka's own design points to force sensing as the control primitive, whichraises tactile content per robot even as it deflates the arm — the one place in the stack where a model-layer winner and its suppliers can both win. The losing position is the integrator who already carries the physical cost of deployment on a 21.6% gross margin and now faces a model layer that can price itself independently of the cell.

What is not disclosed matters as much as what is. Eka has not named a customer, published revenue, or explained how the $150 million funds duty cycles from 5 million to 20 million. Until it does, $1.45 billion is a market convention rather than a measurement — and the honest position is to own the loop's suppliers, not to price the loop itself.