Marcus Cole on the humanoid robotics bet nobody is talking about

Marcus Cole is, by his own admission, the wrong founder for a robotics company in 2026. He's a software person. The category, in his view, is dominated by hardware-first thinking — and that's exactly why he believes Northwind Robotics has a shot.

"The teams beating us on hardware spec sheets are not the teams we're worried about," he told me, from Northwind's Toronto office. "The hardware will commoditize. The teams that own the data and the policy stack on top of that hardware are the ones who win."

The thesis

Cole's view, expanded over the course of the interview, is that the humanoid robotics category in 2026 looks structurally similar to the smartphone category circa 2008. There will be a small number of differentiated platforms, a much larger number of undifferentiated hardware vendors, and the actual long-term value will accrue to whoever owns the equivalent of iOS — the policy layer, the developer ecosystem, the data pipeline.

This is not the consensus view. Most of Cole's competitors are racing to build vertically integrated stacks where they own the actuators, the sensors, the chassis, and the software. He believes that's a category error.

"You can't out-Toyota Toyota. If you bet your company on hardware excellence against incumbents that have been making industrial robots for forty years, you lose. We're not betting on hardware excellence. We're betting on owning the imitation-learning loop better than anyone else."

What that means in practice

Northwind's product strategy reflects the thesis. The company designs its own hardware — Cole rejects the framing that they're hardware-agnostic — but is explicit that the moat is the data infrastructure. Every Northwind unit deployed in the field collects telemetry that flows back into the company's policy training pipeline. The hardware is, in a sense, a data-collection vehicle for the policies.

That positioning has cost the company in fundraising. "There's a particular kind of VC who really wants you to be a hardware company because hardware companies are easier to value. We've had partners walk away from our deck because we wouldn't promise we'd be cap-ex heavy enough."

What he's looking for

The thing Cole is most attentive to right now isn't the next funding round (Northwind raised a $22 million seed in 2024 and Cole says the company has 28 months of runway). It's the question of what the first three industrial customers do.

"You ship a unit to one factory. It does the task. You ship another unit. It does the task. By the third unit, you start learning whether your policy generalizes across environments or whether you've been overfitting to the first deployment. That moment — usually three to six months in — is the actual unit test for the company. We're not there yet. We're close."

For more from our interviews section and our coverage of the humanoid robotics boom, see this section.