Nvidia has put money into software whose selling point is that customers need fewer GPUs. That reads like a company funding its own headwind. In practice it is an attempt to steer demand toward the places Nvidia increasingly cannot serve anyway.
Simplismart, based in Bengaluru, has closed a $14 million Series B. Nvidia joined Accel, Shastra VC and Micromax in the all-primary round, with Dallas Venture Capital leading. The deal values the four-year-old company at about $100 million including the new capital. Amritanshu Jain, formerly of Oracle, and Devansh Ghatak, formerly of Google, founded it in 2022 after meeting at BITS Pilani; it now employs 58 people.
The software sits between a model and the hardware it runs on, adjusting how the model is deployed and managed — response time, request throughput, memory footprint — to find the point where speed, cost and quality line up for a particular customer. Jain frames the thesis bluntly: "Nobody actually needs the fastest or the cheapest inference." What companies need, he argues, is to run models at speeds and costs that make sense for their application, and Simplismart automates the tuning work that otherwise falls to a team of ML engineers. The proof point it offers is InVideo, a video-generation platform that cut GPU serving costs by 56% using the software. HeyGen, Higgsfield and Ema are also customers, and the Indian cloud provider Yotta uses Simplismart to power a serverless inference service.
Growth, by the company's own account, is steep: annualised revenue rose 14-fold in six months to an eight-figure sum. That figure is extrapolated from current customer usage rather than contracted recurring revenue, and excludes money passed through to infrastructure providers. More than half of revenue comes from the United States, despite India having more customers.
The gap between that and the rumour mill is the interesting part. The Economic Times had reported Simplismart in talks to raise $40 million to $50 million at a $300 million to $400 million valuation. Jain declined to comment on the report but said the company was not actively fundraising, that the Series B was closed, and that it would not be extended. Between $100 million and $300 million sits the difference between a solid infrastructure deal and a demonstration that inference optimisation carries as a business in its own right.
Nvidia is not a naive investor here. It already works with the startup on technical integrations around its Dynamo inference software, lends solutions architects and introduces customers. Jain's argument is that efficient inference makes the hardware attractive to more buyers and widens AI adoption overall. That is the industry's standard answer to the efficiency objection, and it is not wrong — it simply does not say what happens if efficiency rises faster than demand.
The segment is already crowded. Nvidia itself, AWS, Google Cloud and Microsoft Azure all offer their own inference paths, and a growing number of startups chase faster and cheaper model serving. Jain sees the niche with customers who face strict SLAs and hard total-cost-of-ownership pressure, not those picking between two presets labelled fast and cheap. The new money is earmarked mainly for US expansion — that is, for the market already paying the bills.
