Nvidia Invests in Simplismart, a Startup That Cut InVideo's GPU Bill by 56%
Nvidia has invested in India's Simplismart. Its software picks the launch settings for an AI model so it responds quickly without tying up extra GPUs. Video platform InVideo used it to cut its GPU server costs by 56%, and the startup's revenue grew 14x in six months. If your bill for running models keeps rising, check the configuration first before buying more hardware.
AI-processed from TechCrunch; edited by Hamidun News
Nvidia has invested in India's Simplismart, a startup that helps companies run AI models on fewer GPUs. Imagine renting expensive servers with GPUs (graphics processors that run neural networks), where half of their time goes to waste because of poor configurations. Simplismart's software picks those settings automatically. According to the company, video platform InVideo used it to cut its spending on running models on GPUs by 56%, meaning it now pays almost half as much for the same responses. The startup's annual revenue, calculated from current client usage, grew 14x in six months.
What the Deal Is
Simplismart closed a Series B round (its third major investment round) for $14 million. Dallas Venture Capital led it, and Nvidia, Accel, Shastra VC, and Micromax also participated. The company is four years old, and with the new money it is valued at roughly $100 million. The round consists entirely of new money: no shares were bought back from earlier investors.
The logic is also interesting. Nvidia sells chips, yet it is investing in a product that helps some customers get by with fewer of them. Founder and CEO Amritanshu Jain explains it this way: when running models gets cheaper, more AI applications get built, and customers buy more Nvidia hardware. Whether this logic holds will be up to the market, but the investment itself says a lot: even a chip supplier sees optimizing model inference as a business in its own right.
How It Works
Simplismart sits between the model and the infrastructure it runs on. Every workload has its own requirements. A chatbot needs a fast reply, batch processing of documents cares more about throughput, and video generation is limited by the cost of an hour of GPU time. Tuning all of this by hand is slow and expensive. The startup's software takes that work over: it chooses how to run a model so that it meets the required response time, requests per second, and compute usage.
Jain frames the approach this way: nobody actually needs the fastest or the cheapest inference (running a finished model to answer users), they need the one that fits a specific product. Competitors, he says, often offer ready-made "faster" or "cheaper" modes, while Simplismart looks for the balance that suits each client. It wins, Jain says, where strict SLAs (commitments on speed and availability) apply and total cost of ownership matters.
The platform can be used as a ready-made service or deployed in a customer's own cloud or on its own servers. Besides running models, it includes fine-tuning, automatic scaling under load, and monitoring. Cloud providers can build their own inference services on it: India's Yotta does this with its Shakti Studio service.
Numbers and Limitations
The company's main argument is InVideo's 56% savings. That figure comes from Simplismart itself, with no independent verification, and the result depends heavily on the client's starting configuration. If a team's infrastructure is already tuned well, the savings will be more modest. Other named clients include HeyGen, Higgsfield, and the enterprise startup Ema.
The 14x revenue growth also needs context. It refers to annualized run-rate revenue, and the final figure is an eight-digit sum; Jain did not disclose the exact number. The calculation is based on recent client usage, not on signed subscription contracts, and it excludes money that passes through to infrastructure providers. More than half of revenue comes from the US, even though there are more clients in India. The new money will go toward expanding in the US, hiring, and developing the software.
The market is crowded. Simplismart competes with AWS, Google Cloud, Microsoft Azure, and Nvidia itself, and alongside them a growing number of startups are making inference faster and cheaper. The company also has a technical partnership with Nvidia: engineers work together, including on integration with the Dynamo inference software. Nvidia also helps with finding customers.
One more detail: The Economic Times reported that the startup was negotiating a $40–50 million raise at a valuation of $300–400 million. Jain declined to comment on that, but said the startup is not raising money right now, that the Series B is closed, and that it will not be expanding.
What This Means for Teams
The practical takeaway is simple. If your product runs models in production, review your inference bill before you order more GPUs. Part of the spending goes to poor configurations: wrong batch-processing parameters, idle GPUs, the same mode used for different workloads. Tools like this should be tested on your own workload, comparing cost per request and response time before and after, rather than relying on the advertised 56%.
- Nov 21, 2026CoreWeave Trains DeepSeek-V3 671B in 17.65 Minutes on Vera Rubin, 2.7x Faster Than GB200
- Nov 21, 2026Huang Steps Down at Nvidia as AMD's Share of AI Chips Grows from 2% to 7–10%
- Nov 21, 2026AssemblyAI cut GPU inference costs by 70% by moving to AWS Spot Instances
- Nov 21, 2026Wavo in Rust Transcribes an Hour of Audio in 18 Seconds, 200x Faster Than Real Time
Key Questions Answered
What the Deal Is?
Simplismart closed a Series B round (its third major investment round) for $14 million. Dallas Venture Capital led it, and Nvidia, Accel, Shastra VC, and Micromax also participated. The company is four years old, and with the new money it is valued at roughly $100 million. The round consists entirely of new money: no shares were bought back from earlier investors. The logic is also interesting. Nvidia sells chips, yet it is investing in a product that helps some customers get by with fewer of them.
How It Works?
Simplismart sits between the model and the infrastructure it runs on. Every workload has its own requirements. A chatbot needs a fast reply, batch processing of documents cares more about throughput, and video generation is limited by the cost of an hour of GPU time. Tuning all of this by hand is slow and expensive. The startup's software takes that work over: it chooses how to run a model so that it meets the required response time, requests per second, and compute usage.
What are the key details on Numbers and Limitations?
The company's main argument is InVideo's 56% savings. That figure comes from Simplismart itself, with no independent verification, and the result depends heavily on the client's starting configuration. If a team's infrastructure is already tuned well, the savings will be more modest. Other named clients include HeyGen, Higgsfield, and the enterprise startup Ema. The 14x revenue growth also needs context.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.