AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Nvidia backs Indian startup helping companies run AI models with fewer GPUs

Collected Oct 27, 2026

Nvidia has taken part in a $14 million Series B round for Simplismart, an Indian startup whose software helps companies deploy and run AI models more efficiently on the GPU infrastructure they already own or rent. The round was all primary capital and was led by Dallas Venture Capital, with Accel, Shastra VC and Micromax also participating alongside Nvidia. The deal puts a valuation of roughly $100 million on the four-year-old Bengaluru company, including the new money.

Simplismart was founded in 2022 by Amritanshu Jain, a former Oracle engineer who is CEO, and Devansh Ghatak, a former Google engineer. The pair met while studying at BITS Pilani and built the company out of their experience building and deploying machine-learning systems. It now has 58 employees.

The product sits between AI models and the hardware they run on. Rather than customers hand-tuning configurations themselves, Simplismart's software adjusts how a model is deployed and managed to hit a customer's targets, optimizing things like response latency, request throughput and compute consumption. Customers can run models through Simplismart's own platform or inside their existing cloud and on-premises environments. The company also offers fine-tuning tools, automatic scaling of compute as demand shifts, and performance monitoring. A separate offering lets cloud providers build and run their own AI inference services on GPU infrastructure using Simplismart's technology.

In practice, inference is the stage where a trained model answers requests, and it is where a large share of ongoing AI spend lands, because GPUs sit idle or underused when batching, memory management and routing are poorly matched to the workload. Software in this layer typically trades latency against throughput against cost, and getting that balance right for a specific application is the problem Simplismart is selling against.

That framing comes directly from Jain, who argued in an interview that no customer actually needs the fastest or the cheapest inference, but rather speeds and costs that fit their particular application. He said the company wins where buyers are stringent about service-level agreements and total cost of ownership, and that unlike rivals offering preset faster-or-cheaper options, Simplismart aims for a balance tailored to each customer.

The company points to AI video platform InVideo, which it says cut GPU serving costs by 56% using its software. Other named customers include AI video companies HeyGen and Higgsfield and enterprise AI startup Ema. Indian cloud provider Yotta uses Simplismart to power the serverless AI inference service inside its Shakti Studio platform.

On the numbers, Jain said annualized revenue run rate rose 14-fold over the past six months to an eight-figure sum, without giving the exact figure. That metric is based on recent customer usage rather than contracted recurring revenue, and excludes revenue passed through to infrastructure providers, the company said. More than half of revenue comes from the U.S., even though the startup has more customers in India. That gap is a familiar pattern in infrastructure software: Indian accounts can be numerous but small, while U.S. buyers tend to run larger workloads and pay more per account.

Nvidia's involvement goes beyond the check. The two companies have been working on technical integrations, including around Nvidia's Dynamo inference software, with Simplismart engineers collaborating with Nvidia solutions architects. Jain said Nvidia has also introduced the startup to potential customers and partners. Dynamo is Nvidia's open-source framework for serving and scaling inference across data-center GPUs, so a startup optimizing model serving has an obvious reason to align with it.

The investment has a surface paradox: Nvidia is the dominant supplier of AI chips, and here it is backing software whose pitch includes running some workloads on fewer GPUs. Jain's counterargument is that more efficient inference makes the hardware more attractive and encourages broader adoption of AI applications, which expands the pie. This is a plausible reading, and it mirrors how chip vendors have long funded software that makes their silicon easier to use. The tension is real, though: efficiency gains that reduce GPU counts per workload put pressure on unit volumes, and the balance only works if cheaper inference pulls in enough new demand to compensate. Whether that holds for Nvidia's customers is an open question, and one worth watching as inference optimization matures.

The competitive field is crowded. Simplismart is up against offerings from Nvidia itself, from hyperscalers including Amazon Web Services, Google Cloud and Microsoft Azure, and from a growing set of startups building tools to make models faster and cheaper to run. Demand in the segment is being driven by rising use of AI applications and the pressure to get more out of expensive GPU capacity.

The new money is earmarked for expanding in the U.S., growing the team and further developing the inference software. Jain said the U.S. remains a key market as the company builds on its existing customer base there.

The announcement follows an Economic Times report that Simplismart was in talks to raise $40 million to $50 million at a valuation of $300 million to $400 million. Asked about it, Jain declined to comment and said the startup was not actively fundraising, adding that the Series B is closed and will not be extended.

Why it matters: Teams running inference at scale now have another vendor claiming to squeeze more out of existing GPU capacity, with third-party figures like the 56% serving-cost reduction at InVideo as the kind of evidence to weigh against hyperscaler and Nvidia-native tooling. The Nvidia backing, and the Dynamo integration work, suggest the chipmaker would rather shape this optimization layer than cede it, which likely means tighter coupling between serving software and its hardware roadmap. For product teams, the practical question is whether workload-specific tuning beats the presets and managed services already on offer, especially where SLAs and total cost of ownership are tight. My inference is that consolidation and partnership pressure in this layer will increase as inference spend becomes the dominant line item in AI budgets.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Simplismart says its annualized revenue run rate has grown fourteenfold in six months, with the U.S. accounting for more than half its revenue.