🔥 SiMa.ai Hit $1.45 Billion Building the Chip Your Robot Runs Without the Cloud
SiMa.ai raised $150M at a $1.45B valuation to run AI on-board robots, drones, and cars — betting the next chip war happens outside the data center.
Every humanoid robot demo you’ve seen this year has a dirty secret: a lot of the actual thinking is happening somewhere else. The robot waves, walks, and pours you a drink, but the heavy AI usually runs on a rack of Nvidia GPUs in a data center, with commands piped back over a network. Cut the connection and the robot gets dumb fast.
A San Jose chip startup just raised a small fortune arguing that arrangement is a dead end.
Let’s get the number on the table first: $150 million. That’s the Series C SiMa.ai closed on September 28, co-led by Fidelity Management & Research and Amplify, which values the company at $1.45 billion and pushes its total funding past $500 million ($150M Series C, $1.45B valuation (TechCrunch)).
The bet behind that check is simple to say and very hard to build: robots, drones, and cars should run their AI on-board, not in the cloud. SiMa.ai wants to be the company that sells them the brain.
🧠Why This Matters
The AI boom you’ve been reading about for three years lives in giant, air-conditioned buildings. It’s centralized, power-hungry, and always online. That’s fine for a chatbot. It’s a problem for a forklift.
Machines that move through the physical world can’t wait 200 milliseconds for a server to answer, and they can’t assume the Wi-Fi holds. They need to see, decide, and act locally, on a power budget measured in single-digit watts. That’s the pitch for “physical AI” — the industry’s new label for intelligence that runs at the edge, inside the device.
SiMa.ai, founded in 2018 by former Groq COO Krishna Rangasayee, builds chips for exactly that job. Its Series C investors — including Dell Technologies Capital, StepStone Group, and Alter Venture Partners alongside the two leads — are wagering that the next wave of AI hardware spending happens outside the data center (Unite.AI).
đź“Š Deep Dive
The product doing the talking is Modalix, SiMa.ai’s system-on-chip built to run multiple AI models — vision, transformers, and language models — on the same low-power silicon. The company pairs it with a software layer called Palette meant to get a model from laptop to robot without a team of chip specialists.
Here’s how the on-board pitch stacks up against the default of shipping data to a GPU in the cloud:
- Compute: Modalix delivers roughly 50 TOPS of multimodal performance at under 10 watts, per SiMa.ai’s specs — the kind of envelope a drone or camera can actually carry (Unite.AI).
- Latency: decisions happen on the device, so there’s no round trip to a server — the difference between a robot that dodges an obstacle and one that logs into it.
- Cost & power: SiMa.ai positions Modalix as the low-power, lower-cost alternative to “expensive, complex, and power-hungry” Nvidia GPUs for edge jobs (BusinessWire).
- Drop-in fit: the module is pin-compatible with Nvidia’s Jetson Orin boards, and a developer kit sells for $1,499 — an aim-at-the-installed-base move (Unite.AI).
- Roadmap: a next-generation part is slated for the first half of 2028, targeting 1,000 dense TOPS in purpose-built physical-AI silicon (BusinessWire).
The customer and partner list is the part that makes investors comfortable: Bosch, Micron, Synopsys, Emerson, TRUMPF, L&T Technology Services, and Alphabet’s robotics-software spinout Intrinsic have all shown up in SiMa.ai’s orbit (BusinessWire).
“While others are still figuring out the pieces or repurposing their cloud offerings, we’ve built the entire puzzle.”
— Krishna Rangasayee, founder and CEO, SiMa.ai
⚠️ The Catch
Read the founder’s own framing and the ambition gets very large, very quickly:
“Physical AI in humanoids, automotive, and drones is the gateway to a $50 trillion market that has remained largely untouched by modern innovation.”
— Krishna Rangasayee (BusinessWire)
A $50 trillion figure is a total-addressable-market flourish, not a revenue line. SiMa.ai hasn’t disclosed how many chips it actually ships or what it earns, and at $1.45 billion it’s valued as a promise, not a P&L — up from under $1 billion just a year ago (TechCrunch).
The competition is also not asleep. Nvidia’s Jetson line owns the edge-robotics mindshare SiMa.ai is trying to pry loose, and Qualcomm, Ambarella, and Hailo are all chasing the same sockets. The headline 1,000-TOPS part that would truly separate SiMa.ai from the pack isn’t due until 2028 — a long time to hold a lead in silicon. And “physical AI” is a phrase doing a lot of fundraising work right now; some of the robots it’s meant to power are still stapled to a demo stage.
🎯 What Happens Next
The $150 million buys time to convert design wins into volume. Watch for two things. First, whether any of those marquee partners moves from evaluation to shipping products with Modalix inside — a named humanoid or ADAS system running SiMa.ai silicon would matter more than any valuation. Second, whether the 2028 roadmap holds; edge-AI buyers plan years ahead, and slipping that date hands the window back to Nvidia.
Expect the “run it on-device” argument to get louder across the industry, and expect Nvidia to answer it. The interesting tell will be pricing: if SiMa.ai can undercut a Jetson on both dollars and watts while matching it on capability, robot makers will listen.
🧩 Bigger Picture
This round is a small, sharp data point in a larger shift. AI’s first act was centralized — bigger models, bigger clusters, bigger power bills. The second act is trying to push intelligence out to the machines that touch the real world, where the constraints are physics, not just budget.
Counterpoint Research projects roughly 145 million cumulative physical-AI device shipments — robots, vehicles, drones — by 2035 (Unite.AI). If even a fraction of that arrives, someone has to supply the low-power brains. SiMa.ai just raised $150 million to be that someone, and it’s far from the only company that spotted the opening. The picks-and-shovels era of physical AI is opening, and the shovels are getting very small.
The cloud made AI think. The next fight is over what lets it move — and whoever wins that ships in the hundreds of millions, one robot at a time.
Sources
