SiMa.ai Raises $150M to Take on Nvidia's Edge-AI Crown

When Fidelity and J.P. Morgan both decide to write checks to the same AI chip startup, the industry pays attention. Neither institution has a reputation for chasing semiconductor hype. Yet both are now on the cap table of SiMa.ai, the San Jose startup that closed an oversubscribed $150 million Series C on September 28, 2026, at a $1.45 billion valuation — and told investors it intends to build the silicon that powers the next generation of humanoid robots, drones and autonomous vehicles.

The round, reported across tech and venture press on September 29 and 30, brings SiMa.ai's total capital raised to $500 million. For a company competing head-on with Nvidia — the most dominant force in AI silicon — it is a significant war chest. For the investors backing it, it is a bet that the most valuable AI compute of the next decade will not sit in a data center at all.

File photo: Tesla's Optimus humanoid robot at the company's October 2024 event in Burbank, California. Humanoid robots are one of the three target markets SiMa.ai is designing its third-generation silicon for. Photo: Steve Jurvetson / Wikimedia Commons, CC BY.

The Money: Fidelity and Amplify Lead a Wall Street-Backed Round

The Series C was co-led by Fidelity Management & Research Company and Amplify Partners. Fidelity is no newcomer here — it has backed SiMa.ai in earlier rounds, including a 2022 Series B extension, making this a double-down rather than a debut.

The more notable signal is who joined. J.P. Morgan, AllianceBernstein, Baron Capital and the State of Michigan all entered as new investors. Fidelity, AllianceBernstein, Baron and J.P. Morgan are large public-market institutions, not traditional venture firms — their presence on the cap table of a pre-IPO chip startup says something about how far "physical AI" has moved into mainstream institutional portfolios.

Existing investors also returned in force: Alter Venture Partners, Dell Technologies Capital, Maverick Capital, +ND Capital, Point72 and StepStone Group all participated.

The round was oversubscribed, according to the company's announcement — a phrase worth noting, because it suggests investors competed for allocation rather than the company scrambling to fill the round.

The new money lifts SiMa.ai's valuation from the $960 million mark PitchBook pegged after its $85 million Series B in July 2025, per TechCrunch's reporting. That is roughly a 50 percent step-up in fourteen months — healthy, not frothy, by semiconductor startup standards.

Who Is SiMa.ai?

SiMa.ai was founded in 2018 by Krishna Rangasayee, who previously served as chief operating officer of chipmaker Groq and, before that, spent 18 years at Xilinx, the programmable-chip company now owned by AMD.

The company describes itself as "software-centric" — a deliberate inversion of how chip startups usually position themselves. Rather than selling raw accelerator hardware and leaving developers to figure out the software, SiMa.ai ships a purpose-built ML system-on-chip (MLSoC) paired with a low-code development environment. The pitch: a robotics or automotive team should be able to take an AI model and have it running on-device in days or hours, not months.

Its funding history traces a steady climb. A $30 million Series A in 2020 was led by Dell Technologies Capital. A $30 million Fidelity top-up in 2022 closed out the Series B, brought total funding to $150 million at the time, and added legendary chip investor Lip-Bu Tan to the board. A $70 million raise in April 2024 funded the first-generation chip's path to market. Now, with $500 million raised overall, the company is operating at a scale few edge-AI startups have reached.

SiMa.ai also has an unusually large engineering footprint in India — roughly 90 people there, with CEO Rangasayee telling BW Businessworld that more than half of the company's global workforce is India-based and that the country contributes 20 to 25 percent of global revenue.

Modalix: What the Silicon Actually Does

SiMa.ai's current hardware line is the Modalix MLSoC family, which entered production in 2025. The chips are built on TSMC's 6-nanometer process and ship in four configurations — M25, M50, M100 and M200 — scaling from 25 to 200 TOPS (trillion operations per second) of AI compute.

The architecture is designed for multimodal edge workloads: four vector processing units for computer-vision algorithms, an integrated image signal processor that talks directly to MIPI and Ethernet cameras, video encode/decode blocks, four lanes of PCIe Gen 5, a 25-megabyte on-chip memory pool, and eight Arm Cortex-A65 CPUs running Yocto Linux for general-purpose tasks.

Crucially, the family supports the full range of modern AI models — convolutional neural networks, vision transformers, large language models and multimodal models — on a single chip. SiMa.ai sells complete systems-on-module, priced at $349 for an 8-gigabyte module and $599 for a 32-gigabyte module at 1,000-unit enterprise volumes, with a $1,499 developer kit.

The company claims the Modalix family delivers roughly 10 times the frames-per-watt of Nvidia's Jetson Orin and Xavier edge chips — a figure it has repeated since the 2024 launch coverage, and one that appears in its partner materials (a joint Synopsys announcement cites a "tenfold increase in performance and energy efficiency"). As with any vendor-supplied benchmark, independent buyers will want to see those numbers on their own workloads. But the direction of travel is clear: SiMa.ai is competing on efficiency, not raw throughput.

The company says its software layer, Palette Neat — an "agentic" development environment for physical AI — is the other half of the equation, and part of the new funding will go toward scaling it. Amplify Partners general partner Mike Dauber said SiMa.ai has worked to reduce the complexity of deploying AI models at the physical edge and shorten the time required to bring autonomous systems to market, calling the company a potential "category-defining technology business."

The 1,000-TOPS Bet: Gen 3 and the Nvidia Comparison

The most ambitious part of the Series C announcement is what SiMa.ai plans to build with it. A third-generation platform is targeted for the first half of 2028, designed to deliver 1,000 dense TOPS of compute on purpose-built physical AI silicon.

SiMa.ai intends to ship the Gen 3 technology in three forms: licensable machine-learning IP, chiplets, and full systems-on-chip. The stated targets are medium- to high-end drones, humanoid robots, automotive driver-assistance systems and AI-powered vehicle cockpits.

Here the company is explicitly measuring itself against Nvidia. Coverage of the announcement describes SiMa.ai positioning the Gen 3 chip as delivering roughly double the performance of Nvidia's Thor platform at an equivalent power envelope, with a 60-watt figure circulating in launch reporting. Treat that framing as the company's claim, not a verified fact — and note the caveats.

Nvidia's highest-end modules, the Jetson AGX Thor series, are rated at up to 2,070 TOPS, but that is a sparse figure measured at 4-bit precision, which counts shortcuts that skip zeros in a model's data. The dense equivalent is roughly half. Thor also draws up to 130 watts at the top end. Comparing across precision formats, sparsity assumptions and power budgets is genuinely tricky, and independent reporting has flagged that SiMa.ai's and Nvidia's headline numbers are not like-for-like.

Still, the technical ambition is real. Moving from 200 TOPS today to a claimed 1,000 dense TOPS in a sub-100-watt envelope by 2028 would put SiMa.ai's silicon in the same conversation as the most powerful edge modules on the market. Whether it gets there on schedule — and whether the numbers hold up outside the company's own testing — is the story to watch.

The Business Case: Why Physical AI Needs Its Own Chips

The funding case rests on a straightforward engineering argument. Humanoid robots, delivery drones and autonomous vehicles all share a constraint that cloud GPU clusters cannot solve: they must make decisions in milliseconds, often with no network connection, inside a thermal budget measured in watts, not kilowatts.

Running those workloads on repurposed data-center GPUs means carrying power consumption, cooling, cost and latency penalties that a battery-powered machine cannot absorb. SiMa.ai argues that purpose-built silicon — designed for sensor fusion, deterministic latency and multimodal processing from the start — is the only way the physical-AI category scales economically.

Counterpoint Research's market numbers back the thesis. The firm projects cumulative shipments of physical AI devices across robotics, automotive and drones will reach 145 million units between 2025 and 2035, with autonomous vehicles the largest segment and humanoids the fastest-growing category — cumulative humanoid installations are expected to pass 100,000 by 2028. Rangasayee has described the opportunity as a multi-trillion-dollar market that modern innovation has barely touched.

File photo: a Waymo robotaxi operating in San Francisco in November 2023. Autonomous vehicles must run perception and planning models on board, with no reliance on cloud latency — exactly the workload SiMa.ai's chips target. Photo: Mliu92 / Wikimedia Commons, CC BY-SA 3.0.

Deal flow is accelerating alongside the technology. Dealroom data cited in recent coverage shows humanoid-robot startups alone raised $8.6 billion in the first half of 2026 — 1.8 times all of 2025. Robot-perception startup Lyte raised $165 million at a $1.6 billion valuation, with Fidelity returning as an investor there too. The capital is clearly flowing toward the physical layer of AI.

On the customer front, SiMa.ai says it serves more than 150 customers, and its materials name Bosch, Emerson, Micron, Synopsys, AVerMedia, ARK Electronics, Kontron, L&T Technology Services, Mistral and STIGA among its partners and clients. Those are company-stated figures and should be read as such — but the breadth of the named list, including multiple Tier 1 industrial and automotive suppliers, suggests the platform has moved past the pure-demo phase.

TechPulse Take: What This Round Really Signals

Analysis: Strip away the funding headlines and this round is really about one question — whether the AI industry's center of gravity is about to shift from the data center to the device.

For five years, edge-AI chip startups have promised a world where intelligence lives on the machine. Most stayed small, selling into niche industrial-vision markets. What changed is the demand side: humanoid robotics went from lab curiosity to an $8.6-billion-a-half funding category, autonomous vehicles are running commercial services across more than a dozen U.S. cities, and drones are moving into industrial inspection at scale. The machines arrived; the power-efficient compute to run their brains is now the bottleneck.

SiMa.ai's bet is that this bottleneck is a silicon problem with a software-shaped answer — that the winner won't be the company with the biggest TOPS number, but the one that makes a robotics team productive fastest. The institutional investor roster suggests Wall Street agrees the category is real, even if the technology race is far from decided.

The honest risks: the Gen 3 roadmap is a roadmap, not a product — H1 2028 is eighteen months out, and semiconductor timelines slip. The Nvidia comparison is company framing built on non-equivalent specs. And Nvidia is not standing still; its Jetson line already spans everything from 400-TOPS mid-tier modules to the flagship Thor.

Still, an oversubscribed $150 million round with Fidelity doubling down and J.P. Morgan, AllianceBernstein and Baron buying in for the first time is a serious vote. SiMa.ai now has the balance sheet to finish the race it started. Whether its silicon can actually double Thor — and whether humanoid robots become a market rather than a demo reel — will determine if the $1.45 billion valuation was a bargain or a mirage. The Series C arrives amid a broader surge of AI megarounds, from Factory AI's recent jump to a $5 billion valuation to billion-dollar robotics raises — and physical AI is where the smartest money is suddenly looking.

Sources

Reporting for this article drew on TechCrunch's September 28 funding coverage; SiliconANGLE's funding and Modalix production-launch reporting; Tech Funding News; TechFlier; The Head and Tale; Ascendants; FourWeekMBA; StartupFeed; SiMa.ai's company materials and product briefs; a Synopsys–SiMa.ai joint announcement carried via PR Newswire/Nasdaq; and historical TechCrunch coverage of the company's Series A and B rounds. Company claims about performance, customer counts and roadmap targets are identified as such in the text.