Nvidia Backs Bengaluru's Simplismart in $14M AI Inference Round

Nvidia doesn't write checks to prove a point. So when the company best known for selling the GPUs that power the entire AI boom puts money into a four-year-old Bengaluru startup, it's worth asking what it sees that the rest of us might be missing.

That startup is Simplismart, and it just closed a $14 million Series B led by Dallas Venture Capital, with Nvidia, Accel, Shastra VC, and Micromax all chipping in. It's the company's second institutional round after a $7 million Series A led by Accel back in 2024, taking its total funding to roughly $21 million.

What Simplismart actually sells

Founded in 2022 by Amritanshu Jain and Devansh Ghatak, both former Oracle and Google engineers, Simplismart works on a problem most AI headlines skip past: it's expensive and slow to actually run an AI model once it's trained. That act of running a trained model to answer a real query — as opposed to the training process itself — is called inference, and at scale it's where companies quietly burn through enormous cloud bills.

Simplismart's software squeezes that process tighter, cutting latency and cost on the same underlying GPU hardware. The flip side of that pitch is just as important: it also helps cloud providers package their raw GPU capacity into services they can actually sell, rather than leaving expensive chips idling.

Why Nvidia is in the room

Nvidia's involvement here isn't just a logo on a press release. Simplismart has been working directly with Nvidia's solutions architects on technical integration, including work tied to Nvidia's own Dynamo inference software. Nvidia hasn't laid out its full rationale publicly, but the pattern is familiar: back the companies building the software layer that makes your hardware easier to use, and more GPUs get bought and rented in the process.

The traction numbers are what likely got investors' attention. The company says its annualised revenue run rate grew 14x over the past six months, with customers including video-AI platforms InVideo, HeyGen, and Higgsfield, along with the customer-support AI company Ema.

"Nobody actually needs the fastest or the cheapest inference," Simplismart co-founder and CEO Amritanshu Jain has said, describing the company's approach as tuning performance and cost to what each customer specifically needs rather than chasing a single benchmark number.

The India angle that matters here

This isn't a foreign story with an India footnote bolted on — Simplismart is an Indian-built company that a global chipmaker chose to back from Bengaluru rather than Palo Alto. That matters because India's AI ecosystem has largely been a consumer of expensive GPU compute, not a producer of the software that makes that compute efficient. Indian AI startups — and increasingly, mainstream enterprises experimenting with generative AI features — routinely cite GPU and cloud inference costs as the single biggest line item slowing deployment. A company solving that problem, with Nvidia's own backing, is a rare case of Indian deep-tech addressing a cost bottleneck the rest of the ecosystem feels directly, rather than building another application layered on top of someone else's model.

According to Jain, the fresh capital goes toward three things: strengthening the inference engine itself, building out the orchestration platform that manages workloads across providers, and expanding into North American cloud partnerships. As he put it, the funding will let the company "take that platform to more enterprises and cloud providers globally."

• $14M Series B led by Dallas Venture Capital, with Nvidia, Accel, Shastra VC and Micromax participating
• Follows a $7M Series A in 2024 led by Accel; total funding near $21M
• Customers include InVideo, HeyGen, Higgsfield and Ema
• Annualised revenue run rate reported up 14x in six months

The bigger pattern worth watching is what this says about where AI spending actually goes next. Training giant frontier models grabs the headlines, but the real, recurring cost for most companies — including Indian ones trying to ship AI features without a Silicon Valley budget — is the everyday grind of inference. Whoever makes that cheaper doesn't need a flashy model of their own to matter.