Nvidia Backs Simplismart's $14M Series B to Cut AI Inference Costs

Nvidia took part in a $14 million Series B round for Simplismart, a four-year-old Bengaluru startup whose software helps companies run AI models more efficiently, in some cases on fewer GPUs, according to TechCrunch.

The round was all-primary, meaning the new capital goes to the company rather than to existing shareholders. Dallas Venture Capital led it. Other Series B investors include Accel, Shastra VC and Micromax. The company was valued at about $100 million including the new funding.

Simplismart operates in the inference layer, the stage where a trained model serves live requests, as opposed to training, where the model is built. According to the same report, the company helped AI video-generation platform InVideo cut its GPU serving costs by 56%. Its customers include HeyGen, Higgsfield and Ema. Yotta, the Indian cloud provider, uses Simplismart's technology to power the serverless AI inference service within its Shakti Studio platform.

Annualized revenue run rate grew 14-fold over the past six months, to an eight-figure sum. More than half of revenue comes from the United States, even though the company has more customers in India. Nvidia has been working with Simplismart on technical integrations, including around Dynamo, its inference-serving software.

Simplismart's previous funding event was a $7 million Series A led by Accel, announced in October 2024 on the company's blog. The new round more than doubles its disclosed capital base at the Series B stage.

The broader context here is the economics of serving models rather than training them. For companies building on open or licensed models, GPU serving cost is a recurring expense that grows with every user request. A reduction of the size reported for InVideo changes which consumer products can be priced sensibly, and how much hardware a customer needs to buy or rent. Software that extracts more throughput from each accelerator sits directly against the cost curve application companies face.

The revenue mix is worth a second look. A company with more customers at home but most of its revenue from the U.S. suggests that willingness to pay for inference optimization is concentrated among American application developers, at least for now. How that balance shifts will indicate whether growth depends on one market or on a product that travels across borders with little adaptation.

The Nvidia relationship deserves weighing on its own terms. Because the reported Dynamo work ties Simplismart's stack to a platform many operators already use, adoption could move faster. It also ties the company's roadmap to software owned by a chip supplier.

In my view, the more useful test for Simplismart is whether its optimizations hold up as the underlying serving frameworks change, since that is where investors and customers will likely judge it.

In my view, what the round makes possible is fairly direct. Cheaper serving lowers the cost of running video generation, avatar and agent workloads at scale, and gives smaller operators access to capacity that once required large infrastructure budgets. Whether Simplismart becomes a durable part of that layer will depend on sustained performance gains, customer retention once the first cost savings are absorbed, and its ability to keep pace with the platforms it depends on.