$55.8 billion in six months.
Nearly double what robotics raised in all of 2025. AI software lost its pricing power. The same funds that gorged on LLMs are now swinging hard at atoms. This is permanent capital reallocation—and it is moving so fast that it mathematically guarantees a brutal sorting when manufacturing reality arrives in 2027.
Consensus says robotics is the next AI wave. It is. But the $55.8 billion headline, per Dealroom, hides a defense-and-infrastructure skew that props up a pre-revenue humanoid sector. The same commoditization that gutted software multiples is already racing toward the hardware layer.
The numbers are not a rounding error
Three data sets. One story.
Crunchbase’s narrower robotics lens catches $18.8 billion through midyear, already above every full-year record. PitchBook clocks over $23 billion, closing fast on the $26 billion captured in all of 2025. Rewind to 2019: the whole space raised roughly $4 billion. The growth curve has gone vertical.
Saronic, an Austin defense-tech firm building autonomous naval vessels, closed a $1.75 billion Series D led by Kleiner Perkins—$2.6 billion in total funding, valuation now $9.25 billion. Skild AI took a $1.4 billion round from SoftBank, tripling its valuation to above $14 billion. Neura Robotics closed a $1.4 billion Series C at a $7 billion valuation, NVIDIA, Amazon, Qualcomm, Bosch, the European Investment Bank, and Tether on the cap table.
Figure AI holds the cumulative crown at $1.75 billion raised. Apptronik sits at $1 billion, UBTECH at $940 million, Agility Robotics at roughly $330 million. Meanwhile, Meta bought humanoid startup Assured Robot Intelligence and folded the team into its Superintelligence Labs unit. Sam Altman declared robotics OpenAI’s next frontier and kicked off hiring that same weekend.
The sector Sand Hill Road once called a hardware graveyard is now commanding infrastructure-sized rounds.
Software can’t charge anymore
Frontier models got roughly 90 percent cheaper. Anyone can build the same AI feature with an API key. About 80 percent of corporate AI deployments show no measurable return. When every startup ships a GPT-wrapper and OpenAI or Anthropic can undercut it at will, software multiples collapse.
Physical AI offers what software surrendered: a supply chain is hard to clone. Real-world training data cannot be scraped. Manufacturing tolerances, regulatory certifications, and the physics of building something that moves create moats a weekend hackathon cannot dissolve. Defensibility now beats capability, and ‘we use AI’ is worth almost nothing.
Jensen Huang, as reported during his GTC Taipei keynote, framed Physical AI as the next frontier—AI that can operate, reason, plan, and act in the real world. The strategic bet is simple: LLMs proved AI can reason; robotics is the wager that it can act at scale, and the same funds are now extending the thesis from bits to atoms.
The kit that commoditizes the robot
Twelve commercial humanoid platforms are buyable, up from three in 2024.
NVIDIA used GTC Taipei to drop a standard humanoid blueprint: a Unitree body, five-fingered hands, NVIDIA onboard computing, and software tools, shipping in late 2026. It’s aimed at academic researchers today, but the signal is unmistakable. The company that commoditized the GPU layer is now commoditizing the humanoid hardware layer.
Neura Robotics’ 4NE-1 hits 0.1N sensitivity and plus-or-minus 0.01mm repeatability. Genuine engineering. But a standard kit from the world’s most valuable chip company turns that precision into a platform feature. The robot builders become vehicle manufacturers in a world where someone else owns the engine and the road. Value migrates to the component suppliers and contract manufacturers, not the assemblers. This is the PC market in 1995, the smartphone market in 2010, the EV market in 2020. The brands that raise the most early capital rarely capture the most long-term value.
$55.8 billion chasing a revenue desert
The headline number is real. It is also dominated by defense and infrastructure plays.
Saronic builds naval vessels for the Pentagon. Skild AI positions itself as a general-purpose intelligence layer that happens to move through hardware, and its valuation reflects an infrastructure, not a robotics, multiple. These are not commercial humanoid companies trying to sell units to factories and warehouses. Strip out the defense-and-platform rounds, and the true market signal looks like the 2021 SPAC pattern in hardware form: enormous capital chases a market that cannot absorb the number of units required to justify the valuations at today’s price points.
The chain that leads to the washout
The funding math guarantees a reckoning. To generate venture-scale returns on nine- and ten-figure rounds, these companies need to sell thousands of humanoid units at six-figure price points—or hundreds of thousands at a lower cost. No factory buyer has validated unit economics at scale. The gap between capital raised and revenue proven is wider than at any point since the autonomous vehicle peak.
Manufacturing reality arrives in 2027. Scaling from prototype to production volume demands massive capex, not R&D. Actuator, sensor, and precision-component supply chains are fragmented and nowhere near the throughput implied by the valuations. The marginal cost to build a humanoid that runs 20 hours a day without failure exceeds what any logistics or manufacturing customer will pay. Picture a procurement officer walking into a meeting late next year with a unit-price quote $50,000 above tolerance: the pilot dies, the backlog evaporates, and the Series D slides into restructuring. These are physics problems. A funding round does not dissolve them.
The consolidation fuse is lit. Within 18 months, at least three top-funded humanoid startups will pivot to defense contracts or be acquired by Big Tech for their hardware teams. The Pentagon is the only buyer with both the budget and the appetite for early-stage reliability. Big Tech needs hardware talent to build the physical layer for its AI ambitions, and acquiring a team at a distressed valuation is faster and cheaper than staffing up internally.
The wrong winners are being celebrated. NVIDIA sells the compute. Sensor-makers sell perception. The small number of factories capable of precision assembly at scale sell production. These businesses earn revenue from every robot built, regardless of whether the robot company survives. The defensible thesis is not the builders but the picks-and-shovels suppliers.
A prediction with an expiration date: By the end of 2028, Figure AI, Neura Robotics, or Skild AI will either land a significant defense contract or be acquired by a Big Tech company. At least one will not survive as an independent commercial entity. If all three are still independent and commercially viable in 2029, this thesis is wrong.
What to do with this information
Don’t speculate on the robot makers unless you see the defense procurement pipeline. Buy NVIDIA. Buy the sensor suppliers. Buy the contract manufacturers that will scale no matter whose logo is on the chassis. Defense robotics companies with active contracts are safer capital than commercial humanoid plays that still run on PowerPoint.
If you’re a robotics founder, get a defense contract or a Big Tech partnership now. The window for raising at these valuations closes the moment one well-funded startup fails a production milestone. The autonomous vehicle playbook is clear: the survivors had government work or a deep-pocketed acquirer who valued the team over the product.
Enterprise buyer? Wait. Twelve platforms today means twelve suppliers desperate for reference customers by 2028. Commoditization pushes prices down. The NVIDIA reference kit, shipping late 2026, accelerates that timeline. First-generation commercial humanoids will depreciate brutally. Let someone else pay for the learning curve.
Watch for the 2027 manufacturing crunch—the first marquee firm to announce a production delay and blame ‘supply chain complexity.’ That is the signal. Distressed assets become available. Hardware teams hit the market. The consolidation wave begins. Position capital before it hits.
Same river, harder current
$240 billion rushed into LLMs and lost pricing power. The same capital is now flooding into robots, dragging the same commoditization dynamics with it. NVIDIA’s blueprint kit is the open-source model of robotics: a standard reference design deletes differentiation at the platform layer and shifts value to whoever controls the components and the lines.
The smart money will be made in the factories. The rest will be named on a cap table and forgotten.
