SiMa.ai closed a $150 million Series C this week, co-led by Fidelity and Amplify with Dell Technologies Capital, J.P. Morgan and AllianceBernstein also writing cheques. The round values the company at $1.45 billion and takes its total funding to $500 million. It makes energy-efficient chips that let drones, robots and vehicles run AI models locally, without shipping data to the cloud — the category now branded "physical AI." It is, by a wide margin, the single biggest funding story to touch Indian AI engineering this quarter.
It is also a Delaware company headquartered on West San Carlos Street in San Jose. India's role in it is a design centre in Bengaluru's Bagmane Tech Park, reporting up to an SVP of Engineering and Operations based in the US. The $1.45 billion of equity value this round just created sits with shareholders of a US entity. The 245-odd engineers building the thing sit, in meaningful number, in India.
That split is not a scandal — it is how semiconductor and deep-tech capital has worked for two decades, and SiMa.ai's founder, Krishna Rangasayee, built it the way every venture-backed hardware company before it was built: US incorporation, US investors, global engineering. But it is worth naming plainly, because the coverage of this round keeps calling it India's physical AI moment, and the honest version is narrower: it is Bengaluru's physical AI engineering moment. The value creation happened somewhere else.
The number India actually controls
Contrast that with what the Indian state is building directly. The IndiaAI Mission — a ₹10,372 crore (about $1.25 billion) program approved in March 2024 — has already deployed 38,000 GPUs at a subsidised ₹65 an hour, well past its original 10,000-GPU target. A third procurement tender now underway adds roughly 3,850 more processing units, including 1,050 Google Trillium TPUs — the first time a non-Nvidia accelerator has entered the Mission's stack. With another 20,000 GPUs confirmed, the public compute pool is on course for 100,000 GPUs by December 2026.
Twelve startups have already been selected to train India-specific multimodal foundation models on that infrastructure. This is the part of the AI buildout where India is the owner, not the landlord: public money, public compute, domestically retained intellectual property, priced to be cheaper than renting from a foreign hyperscaler.
| SiMa.ai Series C | IndiaAI Mission compute | |
|---|---|---|
| Capital | $150 million (private, this round) | ₹10,372 crore / ~$1.25bn (public, since 2024) |
| Who owns the upside | US-domiciled shareholders | Government of India / Indian startups using it |
| India's role | Engineering cost centre | Owner and operator |
| What it produces | A $1.45bn valuation, held abroad | 100,000 GPU-hours of sovereign compute by Dec 2026 |
| IP location | SiMa.ai, Delaware | Twelve India-trained foundation models |
Two different bets, and only one of them pays India back in equity
Neither track is wrong on its own terms. A Bengaluru engineering centre at a $1.45 billion company is a genuinely good outcome for the hundreds of engineers who work there, for the salaries that flow through the local economy, and for the signal it sends about the depth of hardware-adjacent AI talent available in India. Global chip and systems companies keep choosing Bengaluru for a reason, and that reason compounds — more design centres, more senior roles, more of the actual silicon architecture work happening on Indian soil over time rather than just the lower layers of the stack.
But it is a different kind of win from what the IndiaAI Mission is attempting, and conflating the two flatters neither. The Mission's ₹65-an-hour GPUs and its twelve foundation-model startups are a deliberate attempt to own a layer of the AI stack outright — to not be dependent on renting compute or licensing models from somebody else's balance sheet. SiMa.ai's raise, however large the headline number, does not move that needle at all. It is capital and equity accruing to a US cap table, with India supplying the labour that makes the product possible.
The regulatory backdrop is catching up, not leading
Separately, and on a slower clock, India's AI rules are still being written case by case rather than as a single AI law. The IT Rules amendment that took effect in February introduced due-diligence obligations for platforms hosting synthetic or AI-generated content — a deepfakes-and-labelling regime, not a general AI statute. As of July, MeitY officials were still describing a standalone AI regulation as a discussion in progress, with no drafting timeline committed. The ministry's public position, restated multiple times this year, is that there is no reason to slow the AI push and that the regulatory focus should stay on applications rather than the underlying models.
That is a coherent policy stance for a government simultaneously trying to build out 100,000 GPUs of sovereign compute capacity — you don't throttle the thing you are racing to build. But it leaves India's actual AI-specific rulebook well behind the capital and compute numbers moving every week, this one included. The compute is catching up to the ambition faster than the regulation is catching up to the compute.
The domestic funding number that gets lost next to SiMa.ai's
Indian AI startups themselves are not short of capital right now — they raised close to $1.07 billion in the first half of this year alone, up roughly a third on the same period last year, with deal count climbing from 112 rounds to 157. That money is overwhelmingly going into agentic software, enterprise tooling and consumer applications built on top of somebody else's foundation models, not into chips or physical-AI hardware platforms of SiMa.ai's scale. It's real, it's growing, and almost none of it produces an equity stake as large as the one Fidelity and Amplify just wrote into a single round this week.
That asymmetry is the actual takeaway, more than any single company's cap table. India is simultaneously the fastest-growing source of AI application funding by deal count, the owner of one of the largest sovereign GPU build-outs in the world, and still a bystander to the biggest physical-AI valuation event touching its own engineering talent this quarter. All three things are true, and none of them cancel the other two out.
