Why Nvidia is investing in Simplismart, a startup that helps customers use fewer GPUs

TechCrunch reports that Bengaluru-based Simplismart raised a $14 million Series B led by Dallas Venture Capital, with Nvidia, Accel, Shastra VC and Micromax participating. The round values the four-year-old company at about $100 million. Simplismart was founded in 2022 by former Oracle engineer Amritanshu Jain and former Google engineer Devansh Ghatak, and has 58 employees. Its software sits between AI models and the hardware they run on, tuning response speed, throughput and compute use so companies get more from the GPUs they already have. Nvidia has worked with it on technical integrations, including its Dynamo inference software.

Simplismart says it helped AI video platform InVideo cut GPU serving costs by 56 percent, and counts HeyGen, Higgsfield and Ema as customers. Jain says annualized revenue run rate grew 14-fold in six months to an eight-figure sum, based on recent usage rather than contracts. More than half of revenue comes from the U.S., and regional clouds such as India’s Yotta can build serverless inference on its software.

Nvidia backing software that lowers GPU needs makes sense if cheaper inference brings more AI applications into existence, and those applications still run on Nvidia hardware. It’s the old efficiency story: make something cheaper to use and people tend to use more of it. Nvidia has made the same bet at larger scale, paying $12.93 billion for Hugging Face to anchor where open models live (our analysis).

The product idea is also right. Jain’s line that nobody actually needs the fastest or the cheapest inference gets at something true: most businesses need a specific balance of speed and cost for a specific job, and tuning that by hand is hard, slow and expensive. Inference, running a trained model to answer requests, has become a major share of compute spending as AI products scale. OpenAI’s own work speeding up GPT-6 Astra on Blackwell shows how much the big labs invest in the same problem (coverage).

A 14-fold run-rate jump based on recent usage isn’t contracted recurring revenue, and the exact figure wasn’t shared. The 56 percent savings comes from a single customer case. The market is also crowded: Nvidia itself, AWS, Google Cloud, Azure and many startups are chasing inference optimization. A small company needs to keep proving it beats tools its own investor sells.

Gains like this can ease pressure on GPU supply and power. They don’t reduce total compute demand if efficiency fuels more usage, and the history of efficiency in tech suggests that’s the likelier outcome. Cheaper inference is good for builders; it isn’t an answer to AI’s energy footprint. That tension is playing out in data center fights from Memphis (the xAI turbine case) to grid-flexibility efforts. More in our startups coverage, funding coverage and NVIDIA coverage.