In early 2024 our family office made its first meaningful allocation to robotics — a $35M commitment spread across three specialist funds and four direct co-investments. Two years in, we are net down on paper by about 11%. We also believe it is the best allocation we have made in the last five years. Here is why those two sentences do not contradict each other.
The J-curve is real and it is steeper than you expect
Every LP knows about the J-curve in theory. In hardware robotics specifically, the shape is more pronounced than in most of venture. Hardware companies burn meaningful capital before they ship meaningful product, and unit-economics-positive revenue typically arrives two to three quarters after the round that was supposed to get them there.
The result is that a 2024 robotics vintage will often look worse at month 24 than a 2024 SaaS vintage at the same point in time. It does not mean the investment is worse. It means the pattern of markup is different. First-time LPs who do not internalize this in advance often panic at exactly the wrong moment.
Reserve strategy is the difference between returns and failure
A $10M commitment to a robotics fund, with no reserve discipline on top, is a commitment to underperform. The best robotics funds we have worked with ask their LPs to reserve an additional 50–80% of the primary commitment for pro-rata in the best performers. That is a large ask. It is also what separates the 2x DPI funds from the 1x DPI funds in this sector.
The robotics funds with the worst realized returns were not the ones that picked the wrong companies. They were the ones whose LPs did not show up for the pro-rata when the winners emerged.
Our rules after two years
- Treat the primary commitment as 60% of the total capital to the strategy, not 100%
- Pre-commit the pro-rata reserves in the governance memo, not mid-cycle
- Cap total sleeve exposure at 12% of the portfolio — hardware beta is real
- Avoid blind-pool commitments to generalist funds for robotics exposure
- Require quarterly fleet-level data from co-investments, not just financial reporting
The hardware beta is worth the discomfort
The reason we stayed in after the first twelve-month markdown is simple: the asset class has a structural tailwind that pure software does not. Automating physical work is a larger economic opportunity than automating information work, and the window in which capital can participate in the formation of that market is narrow.
What I would do differently
I would make a smaller primary commitment and reserve more aggressively. I would weight the fund selection toward managers who had personally run robotics operating roles, not just invested in robotics companies. And I would refuse any co-investment opportunity where the GP could not produce six months of field-level telemetry from the underlying fleet.
The 11% mark-to-market loss two years in is, in hindsight, almost exactly what I should have modeled going in. The mistake was not the allocation — it was the expectation. This is a category where patience is genuinely compensated, and impatience is genuinely punished.