Why Physical AI Needs Specialist Capital

The money is already moving, and faster than most people realise. Robotics and physical AI startups raised on the order of $27.6 billion in 2025 across more than 1,000 deals (PitchBook), more than double 2024, and 2026 is running hotter still: by mid-year the sector had already crossed $23 billion, on track to pass all of last year, with humanoid rounds setting the pace and Tesla committing $20B in Optimus capex for 2026. Figure, a single humanoid company, closed more than $1 billion last September at a $39 billion valuation (Figure), roughly fifteen times what it was worth eighteen months earlier. In Europe, Neura Robotics just raised $1.4 billion at a valuation of about $7 billion, making it one of Europe's best-funded AI startups. In Silicon Valley, AI took 93 percent of every venture dollar in 2025, $103.5 billion out of $111 billion (Mind the Bridge "Silicon Valley Bets Big on Physical AI”). Robotics, humanoids and embodied intelligence have become the new front of the AI gold rush.

The pattern is familiar. Generalist mega-funds are doing in robotics what they already did in software AI: crowding into the most obvious deals, at the highest prices, on the shortest timelines. Simon Lancaster makes the point in Unlocking Alpha: The Rise of Niche VC, where he argues the industry has “traded away focus for breadth and conviction for consensus. The result is a model that struggles to generate true alpha.”

Robotics is where that model breaks.

Robotics runs on different physics

Software-AI investing rewards speed. Back the team with the best model, scale the GPU bill, race to product-market fit. Robotics does not work that way. A humanoid that climbs a flight of stairs in a demo is a long way from a fleet that runs a warehouse for a year with nobody stepping in. Sensors, actuators, training data, certification, supply chains: every one of them needs slow, patient engineering. We have already seen where the shortcuts lead. After more than $100 billion poured into autonomous vehicles, with only a handful of US cities — San Francisco, Phoenix, LA, Austin — running fully driverless commercial service today.

The problem is not generalist capital itself. It is a mismatch of capital to the stage. When money that is priced for software speed sets the tempo for robotics, three things tend to go wrong. Valuations drift away from any real technical milestone. Founders skip the messy learning years and jump straight to deliver-a-moonshot-or-die. And the early cap table fills up with investors who are there for the story more than the substance, and who cannot help much when the hardware misbehaves.

It is not only the headline rounds. We watch early-stage robotics teams quietly rewrite their decks to ride the foundation-model wave, trading a credible story about a new gripper or perception stack for a vague line about being “the OpenAI of robotics.” That pivot might close the round. It rarely builds the company.

What specialist VC actually looks like

The fix is not less capital. It is better-matched capital. Frontier hardware needs investors who can price a deal because they actually understand the physics, the supply chain and the competitive map. Investors who can sit still through a twelve-month detour into a new actuator design, because they have lived through that detour before. Investors who know which Tier 1 OEM to call for a pilot, and which one to keep well away from.

Lancaster and Sabrina Paseman put it well: “Niche VCs have an edge because they aren’t trying to be everything to everyone. Founders can tell when an investor deeply understands their space and can plug them into a relevant network immediately. That kind of credibility wins trust, and access, faster than brand alone.”

This is how we work at FOV. We are a specialist fund backing European founders building for a new era of computing, one that is more intelligent, more spatial and more human. The convergence of AI, robotics and spatial computing has been our mandate since 2016. Across Fund I we have backed 33 companies on that thesis, and the physical AI names are where the specialist edge shows up most clearly.

To be clear, that edge is most powerful when it is paired, not isolated. We like to lead or co-lead the pre-seed, where deep domain knowledge does the most work. We also like co-investing as the specialist alongside a larger generalist fund, where we bring the technical conviction and network and they bring the firepower to scale. Some of the best cap tables in this category have both.

A few of them:

Makiina (Finland) is building lower-cost robotic arms for industrial use. Not a viral demo, a serious run at the cost curve in a market a few incumbents have owned for decades. We led at pre-seed while most generalists were still asking whether industrial robotics was “AI enough.”

Spogen AI (Finland) is bringing AI to heavy machinery, the tractors, harvesters and excavators that actually move the physical economy. Hard problem, fragmented buyers, long sales cycles. Exactly the kind of company that gets overlooked when capital is chasing the obvious narrative, and that a specialist syndicate can build patiently.

Levtek (Sweden) is developing ride-on cognitive robots that reason, learn and adapt, putting AI to work in physical labour. Embodied intelligence for real workplaces, not stage demos.

Distance Technologies (Finland) is turning windscreens and aircraft canopies into mixed-reality displays, the optics and perception layer physical AI needs to operate inside vehicles. GV led the seed round, following the conviction we had backed earlier.

None of these fits a one-line trend. They fit a long bet on the layer where atoms meet algorithms. That bet only pays off if the early investors can price it, support it and stay in the room when the work gets hard.

Why this matters for LPs

For limited partners, the physical AI cycle is the cleanest test of specialist-versus-generalist returns we have seen in years. The shape keeps repeating: a handful of mega-rounds wrapped in billion-dollar narratives, and a long tail of category-defining companies that need patient, expert money to get built at all. Underwriting the heroic version of those mega-rounds, at scale, is a lot of LP capital riding on a binary outcome.

The LPs we are talking to for Fund II keep telling us the same thing in different words. They are over-exposed to generalist funds chasing the same logos. They want specialists who see what others miss, who hold price discipline in hot markets, and who know the difference between a robotics company that should raise $5 million and one that genuinely needs $50 million. Niche is becoming a core allocation, not a hedge.

For founders the message is simpler. Money is fungible. Partner value is not. When the prototype dies at 2am, the investor who knows the field beats the one with the biggest logo.

For our peers in the generalist tier, none of this is zero-sum, and we do not treat it that way. Several of them are partners we co-invest with happily, and some are investors in our own funds. Mega-funds will and should power the growth rounds once a physical AI company is ready to scale. The point is narrower than “generalists versus specialists.” It is that the earliest rounds in physical AI reward a particular kind of expertise, and the strongest syndicates pair that expertise with the capital to scale later.

Where this leaves us

Frontier tech is too important to fund on reflex. The next decade of physical AI and robotics will be built by founders who go deep, and by investors who know how to back them at a sensible pace and a sensible price. Focus beats frenzy. That is the bet we are making with Fund II, and we think the next generation of European hardware-and-AI champions will prove it out.

Reference: Unlocking Alpha: The Rise of Niche VC by Lancaster, Paseman, Felix Staeritz, Michael Ströck (Amazon).

Viewpoints is brought to you by FOV Ventures, the leading European fund investing in the next era of computing.

LinkedIn | Website | Pitch us