AI Models Are Getting Bigger. Simplismart Is Betting the Bottleneck Is How They Run

Training an AI model is only part of the cost equation. Once that model reaches production, every response, video, voice output or multimodal request consumes expensive compute. Simplismart has raised $14 million in Series B funding to make that inference layer faster, cheaper and easier to operate at scale.

The round was led by Dallas Venture Capital, which invested close to $5 million, with participation from NVIDIA, Accel, Shastra VC, Micromax and angel investors. The fresh capital will support product development, partnerships with cloud and sovereign-AI providers, and expansion across North America and other global markets.

Founded in 2022 by Amritanshu Jain and Devansh Ghatak, Simplismart builds the software layer between AI models and GPU infrastructure. Its inference engine lets enterprises tune how models run based on cost, latency and performance requirements, while its orchestration platform handles deployment and scaling across cloud and on-premises environments.

The next major piece is dynamic GPU scheduling. Simplismart wants workloads to move toward available compute capacity across regions and time zones instead of forcing every customer onto a fixed infrastructure setup. For cloud providers, its software can also turn existing GPU capacity into managed AI services under the provider’s own brand. Yotta is already using Simplismart for the serverless inference layer inside its Shakti Studio platform.

The startup supports shared endpoints, dedicated endpoints and bring-your-own-compute deployments, with the latter aimed particularly at enterprises that need tighter control over data and infrastructure. Customers include HeyGen, Higgsfield, InVideo, Shiprocket, Apna and Ema across video, voice, multimodal and enterprise AI workloads.

Simplismart says its annualised revenue run rate has grown 14X in the past six months, while about 55% of revenue now comes from the US. The company’s larger bet is that AI infrastructure competition will increasingly shift from who owns the most GPUs to who can make existing GPUs deliver more useful work.