The AI Chip Market Everyone Forgot: Robots That Don't Talk

Published: October 2, 2026 | Category: Technology | By Mahesh

Today's funding news is SiMa.ai, which closed a $150 million Series C this week, pushing its total capital raised to $500 million and its valuation to $1.45 billion [1]. The round is notable mainly for what SiMa.ai is not building. Every major chip story this year, from Nvidia's data center backlog to the CoWoS packaging bottleneck Depth Grid covered in September, has been about the chips that run large language models in the cloud. SiMa.ai builds something else entirely: chips designed to run AI inference directly on humanoid robots, cars and drones, out in the physical world, with no data center connection involved at all [2].

The company describes its focus as "Physical AI," a category distinct from the generative AI boom in a specific technical way. A chatbot can tolerate a round trip to a cloud server and a second or two of latency; a humanoid robot deciding whether to put its foot down on a step cannot. Inference for a physical system has to happen on the device itself, in milliseconds, on a power budget a data center GPU was never designed to meet. That is the market SiMa.ai is building for, and it is a genuinely different competitive landscape than the one Nvidia, AMD and the hyperscalers are fighting over.

Naming humanoids, automotive and drones together as the target customer base is itself a bet on convergence. Those three categories have historically run on different chip stacks built by different suppliers, automotive chips need safety certification regimes that drone chips don't, and humanoid robotics is young enough that no dominant chip standard has formed yet. SiMa.ai's pitch is that the underlying problem, real-time perception and decision-making on a power- and heat-constrained device, is similar enough across all three that one chip architecture can serve all of them, rather than each vertical needing its own bespoke silicon.

What to Watch Next

The real test for this round is whether SiMa.ai can land a named design win at scale in at least one of its three target markets before the next funding cycle, since physical AI chip startups tend to live or die on landing one anchor customer that proves the architecture works outside a lab demo. Worth watching alongside this: whether Nvidia, which already sells Jetson edge chips into robotics and has been circling the humanoid space through its own Isaac platform, responds by cutting into this market directly rather than leaving it to smaller specialists.

Sources

1. Business Wire, "SiMa.ai Reaches $1.45B Valuation with $500 Million in Total Funding to Scale Physical AI in Humanoids, Automotive, and Drones," October 2026. businesswire.com
2. Business Wire, "SiMa.ai Reaches $1.45B Valuation" (see source 1).

Read More on Depth Grid

• The Real AI Chip Bottleneck: CoWoS Packaging
• Ask a Company How Many AI Agents It's Running. Watch It Shrug.
• Enterprise AI Agents: Where the Big Checks Go
• Google ADK for Kotlin 1.0: AI Agents Running On-Device
• OpenAI Shelved Its Next Model Over Lying. Then Shipped Agents on the Old One Anyway.

Article by Mahesh | Depth Grid