Analysis
Three sectors that have nothing else in common -- general venture funding, defense tech, and physical AI -- are producing the identical shape of chart in 2026: dollars up sharply, deal count barely moving. Global venture investors deployed $42 billion in August across more than 1,500 startups, more than double August 2025's total even as the number of companies funded stayed roughly flat with prior months. Defense tech startups have already cleared 2025's full-year funding record with a third of the year still to go, on a pace of 107 rounds that is only modestly above last year's volume. Robotics and physical AI companies have raised tens of billions in 2026 according to Dealroom, a figure that would have been unthinkable three years ago. None of these are stories about more companies getting funded. They are stories about the same or slightly more companies getting funded much larger checks.
The mechanism is easiest to see in the seed stage, where the unit of measurement has stopped resembling a seed round at all. Pulse has tracked at least two AI labs -- Ineffable and Thinking Machines -- that raised a billion dollars or more in rounds their own investors called seed financings, at valuations of $5.1 billion and $12 billion respectively, before either had shipped a product or booked a dollar of revenue. Thinking Machines is reportedly now in talks for a new round at a steep markup, a multiple-x jump in roughly six months. That kind of markup used to require years of revenue growth to justify. Now it requires a compute roadmap and a recognizable founding team.
Defense tech tells the same story with different actors. A Crunchbase count puts total 2026 deal volume at 107 rounds, barely ahead of the roughly 106 rounds logged across all of 2025, even as the category's dollar total has grown by more than 50% year over year. Capital isn't spreading across more defense startups; it is stacking on top of the handful that already won their first fights for institutional attention:
“- XPENG physical-AI unit -- over $900M raised at a $6.3B valuation to scale its IRON humanoid platform (China-based).”
- Anduril -- $5B Series H at a $30.5B valuation: autonomous weapons and counter-drone systems.
- Castelion -- largest recent single defense round: hypersonic missile systems, a multi-year hardware development cycle most venture investors avoided until this year.
Robotics shows a milder version of the same curve, in part because there are more credible physical-AI companies to spread capital across, but the check sizes tell the same story:
- [Atoms](/pulse/atoms-1-7-billion-robotaxi-a16z-2026) -- Travis Kalanick's robotics company, $1.7B Series A led by Andreessen Horowitz, $100M from Uber.
- Neura Robotics -- Series C worth up to $1.4B at roughly a $7B valuation, backed by Nvidia, Amazon, Qualcomm and Bosch.
- XPENG physical-AI unit -- over $900M raised at a $6.3B valuation to scale its IRON humanoid platform (China-based).
What ties robotics to defense and to the general venture numbers is not that fewer companies are getting funded -- it's that the size of a competitive round has reset upward across every category simultaneously, and it happened fast enough that no single company's fundraising history explains it on its own.
Why this reads as one pattern, not three
Compare August's $42 billion total to July's number and the picture gets more interesting: the month-over-month figure actually fell 25%, even as the year-over-year number more than doubled. That combination -- a huge YoY gain riding on a handful of outsized months rather than a smooth uptrend -- is what concentration looks like in aggregate data. A small number of mega-rounds (Databricks' round was August's largest, at $5 billion) can swing an entire month's global total on their own, which means the headline growth figure is now much more sensitive to which seven or eight deals happen to close in a 30-day window than it was two years ago, when capital was spread across a wider base of mid-sized rounds.
The counterweight here is real: announced capital and deployed capital are not the same thing, and several of these headline rounds -- Thinking Machines' reported $40 billion talks chief among them -- have not actually closed. A number of these companies are also multi-year build-outs (robotics and defense hardware in particular) where the capital gets drawn down over 24-36 months rather than spent immediately, so the dollar totals overstate how much economic activity is actually happening in any given month. And concentration into fewer, larger bets is not obviously worse for returns than the old spray-and-pray seed model -- it may just be a rational response to compute and hardware costs that have themselves concentrated around a handful of suppliers.
For founders outside the small set of category leaders, the read is uncomfortable: the same dollars that used to fund five $10 million seed rounds are now funding one much larger one, which raises the bar for what a first check needs to look like to get attention at all. For LPs, the honest question is whether a portfolio built this way behaves like venture or like a concentrated public-market bet dressed in venture terms -- a small number of managers writing large checks into a short list of the same category leaders everyone else is also chasing. Watch whether September's numbers show the same shape once Thinking Machines' reported new round, if it closes, gets counted.
