Bengaluru AI infrastructure startup Simplismart has raised $14 million in a Series B round led by Dallas Venture Capital, with NVIDIA, Accel, Shastra VC and Micromax taking part.
The company announced the round on October 27, 2026. It will use the money to develop its inference engine and orchestration platform, expand its cloud provider partnerships and grow across North America and other global markets.
The software between the model and the GPU
Simplismart builds the software that sits between AI models and GPU infrastructure, managing deployment, scaling and day to day operation so that each workload meets its performance and cost requirements. Its modular inference engine adjusts serving settings for latency, throughput, quality and cost, rather than using one fixed setup for every workload.
That tuning shows up in real customer numbers. For enterprise AI company Ema, the target was about 200 milliseconds of latency at roughly 9 cents per million tokens. Simplismart delivered a production setup at 7 cents and 180 milliseconds, after testing cheaper and faster options on either side.
Why NVIDIA is backing an efficiency play
The most striking name on the cap table is NVIDIA, the company whose chips power most AI workloads. Simplismart works with NVIDIA on technical integrations, including work around its Dynamo inference software, and NVIDIA has introduced the startup to potential customers and partners.
Simplismart has also documented its NVIDIA journey on video.
There is an apparent tension here, since software that runs models more efficiently can mean fewer GPUs per workload. Jain has argued the opposite effect wins out in the end, because cheaper and more reliable inference makes AI applications more attractive and encourages wider adoption.
The competitive field is crowded. Simplismart faces inference offerings from NVIDIA itself and from major cloud providers including Amazon Web Services, Google Cloud and Microsoft Azure, alongside a growing set of startups building tools to run models faster and cheaper.
Jain has said Simplismart tends to win where customers face strict service level agreements and tight cost constraints. That is a revealing claim, because it positions the company not as the cheapest option on paper but as the one that holds up when a production system has to meet a promised standard.
With frontier-class open models now readily available, access alone is no longer enough to set a business apart. The advantage comes from how you run them: what each interaction costs, how quickly the application responds, and whether it stays reliable as demand grows.
Amritanshu Jain, co-founder and chief executive of Simplismart, said this as the company announced the round.
A business built on usage
Simplismart sells through three models. Shared endpoints charge by token use with no minimum fee, dedicated endpoints rent GPU infrastructure bundled with its optimization software, and a bring your own compute model lets regulated businesses such as banks run workloads on their own infrastructure.
Growth has been sharp. The company says its annualised revenue run rate has grown 14 times over the past six months, driven by customer usage across video, voice, multimodal and enterprise AI workloads. Around 55 percent of revenue comes from the United States, with the rest from India.
From BITS Pilani to a global customer list
Simplismart was founded in 2022 by Amritanshu Jain and Devansh Ghatak, who met while studying at BITS Pilani. Jain previously worked at Oracle Cloud AI on language services for healthcare, while Ghatak worked as a machine learning engineer at Google. The company now employs 58 people.
The startup began as a platform to help teams build and manage machine learning systems, and it was picked for the AWS ML Elevate programme in 2023. It raised seed funding from Anicut Capital, Sunn91, First Cheque and Shastra VC, then a $7 million Series A led by Accel in October 2024.
Its customer list now spans video, voice and enterprise AI. HeyGen, Higgsfield, InVideo, Shiprocket, Apna and Ema run production workloads on the platform, and Indian cloud provider Yotta uses Simplismart to power the serverless inference layer of its Shakti Studio platform.
What the money funds next
Beyond the core platform, Simplismart is working toward a dynamic GPU scheduling layer that matches demand with available capacity across regions, time zones, workloads and compliance requirements. Idle capacity in one region could eventually serve demand elsewhere when timing and requirements line up.
The round leaves Simplismart with a notable backer in NVIDIA and a clear test ahead. The next proof point is whether its efficiency edge keeps converting into production workloads as enterprises move from AI experiments to systems that must run reliably every day.
