Analysis
I think the venture industry is mispricing what robotics capital actually is right now, and it's going to matter for how LPs think about returns from this category over the next five years. Robotics and physical-AI companies have raised $55.8 billion in 2026 according to Dealroom -- a number that gets reported and discussed like it's evidence of a normal, if unusually hot, venture cycle. It isn't. Look at where that capital is actually landing and it reads much more like project-finance infrastructure spending wearing venture-round terminology.
Start with the check sizes and the backers:
- Neura Robotics -- Series C worth up to $1.4B, backed by Tether, Qualcomm, Amazon, Nvidia, plus industrial giants Bosch and Schaeffler and the European Investment Bank. An EIB commitment is about as close to project-finance infrastructure capital as you'll find in a venture round.
- [Atoms](/pulse/atoms-1-7-billion-robotaxi-a16z-2026) -- $1.7B round Andreessen Horowitz is calling a Series A, with Uber putting in $100M directly as a strategic.
- XPENG physical-AI unit -- over $900M raised at a $6.3B valuation to scale a single humanoid robot platform (China-based).
“- Atoms -- $1.7B round Andreessen Horowitz is calling a Series A, with Uber putting in $100M directly as a strategic.”
These aren't seed-to-Series-B venture bets on an uncertain market; they're capital-intensive build-outs of specific hardware programs with multi-year development and manufacturing timelines, funded by a mix of strategics, sovereign-adjacent capital, and development-bank-style institutions that don't behave like traditional venture LPs.
That mix of backer matters because it changes what "return" means for the capital actually flowing into the category. A traditional venture fund needs power-law outcomes -- a small number of massive winners covering the losses on everything else -- within a roughly 10-year fund life. Qualcomm, Amazon, Bosch and the EIB aren't optimizing for that. They're optimizing for strategic access, supply-chain positioning, or policy-aligned industrial development, on time horizons that can run considerably longer than a venture fund's life. When that capital shows up inside a round venture funds are also participating in, the pricing it sets doesn't necessarily reflect what a venture-only investor should be willing to pay for the same equity.
Room for disagreement: robotics bulls would point out that every capital-intensive category eventually produces its own venture-scale winners -- cloud infrastructure looked like a capex-heavy, infrastructure-adjacent bet in 2010 before AWS, Azure and Google Cloud became some of the best venture-adjacent returns of the decade, and Tesla itself spent a decade looking more like a project-finance story than a software one before it didn't. It's a fair point, and the honest answer is that nobody knows yet whether physical AI resolves the same way. But cloud infrastructure had a clear, fast-scaling unit-economics story within a few years of the first big rounds; robotics hardware companies are still mostly pre-revenue-at-scale, with unit economics that depend on manufacturing learning curves nobody has proven yet at the volumes these valuations assume.
What I'd actually do with this: treat participation in the largest robotics rounds as adjacent to infrastructure investing, not classic venture, and size positions and return expectations accordingly rather than assuming a normal venture power-law outcome bails out a high entry price.
