Simplismart: $14 million Series B funding, NVIDIA invests

Simplismart has announced a $14 million Series B funding round led by Dallas Venture Capital, with participation from NVIDIA, Accel, Shastra VC and Micromax, according to the AI infrastructure company’s funding announcement. Simplismart said it will use the money to develop its inference software, expand cloud partnerships and grow internationally, with North America a priority.

Dallas Venture Capital said it is investing $5 million in the round. Amritanshu Jain, Simplismart’s co-founder and chief executive, told Moneycontrol that the $14 million raise is a standalone funding round, not the first instalment of a larger investment.

The Economic Times identified Simplismart as a Bengaluru-based AI company. It builds software for AI inference: running a trained model to generate responses or process information. Simplismart said its software helps businesses meet cost, response-time and deployment requirements when their models move into production.

Software rather than a GPU fleet

Simplismart said it has chosen to build software that turns third-party graphics processing unit, or GPU, capacity into AI services rather than owning a GPU fleet. GPUs provide the computing power used to run these workloads.

The company described two customer groups. Application businesses use its platform to operate their AI products, while cloud providers use the software to offer managed AI services under their own brands.

Simplismart’s inference engine configures how models serve requests. Its orchestration platform coordinates deployment and manages capacity as demand changes, according to the company. Together, these products handle the infrastructure work while leaving customers control over deployment choices, Simplismart said.

Dallas Venture Capital said the platform works across different cloud environments and allows enterprises to fine-tune, deploy and monitor generative AI models on their own infrastructure. Fine-tuning adapts a model to a particular task or dataset.

The investor said Amritanshu Jain and Devansh Ghatak founded Simplismart in 2022. The company previously raised a $7 million Series A led by Accel in October 2024, with participation from Shastra VC, Titan Capital, Dallas Venture Capital and Notion co-founder Akshay Kothari, according to Dallas Venture Capital.

Growth figures measure different things

Simplismart said its annualised revenue run rate grew 14-fold over the six months preceding its announcement. The Economic Times also reported that Jain gave it that figure.

An annualised run rate projects a recent revenue pace over a year; it is not the same as revenue earned during a completed financial year. Simplismart’s announcement does not give the starting or ending revenue amount behind the growth multiple.

Moneycontrol separately reported Jain as saying revenue had grown tenfold “this year”. That statement and the company’s six-month run-rate figure use different descriptions and reporting periods. They are not interchangeable measures.

Jain also told Moneycontrol that around 55 per cent of revenue came from the US, despite Simplismart having more customers in India. That revenue mix provides context for the company’s stated North American expansion priority.

Simplismart named HeyGen, Higgsfield, InVideo, Shiprocket, Apna and Ema as businesses running production workloads on its platform. The company said those workloads span video, voice, systems handling multiple types of input and enterprise AI.

The cost and response-time trade-off

Simplismart’s announcement explains its technology through the different demands placed on AI applications. A voice agent needs a quick response to maintain a conversation, whereas an overnight document-processing task can accept a longer wait in exchange for lower costs.

The company said configuring a workload involves decisions about hardware, numerical precision, batching, caching and scheduling. Batching groups requests for processing together; larger batches can increase the amount of work completed while making individual requests wait longer.

Reducing numerical precision can lower computing requirements but affect output quality, Simplismart said. Its stated objective is therefore not to maximise a single benchmark, but to find a configuration that meets an application’s combined requirements under actual traffic.

For voice AI provider Exotel, Simplismart reported managed inference costs of Rs 0.90-Rs 1.30 per minute, compared with Rs 2.10 per minute for a closed-model application programming interface setup. An application programming interface, or API, lets software send requests to another service.

Simplismart described the cost reduction as approximately 38-57 per cent in that workload comparison. It also reported latency, or response delay, below 300 milliseconds, against 1.1 seconds for the closed-model setup.

These are Simplismart’s reported results for the Exotel comparison, rather than a performance guarantee across models or customers. The company said Exotel had further trained an open-source model on its own data. Allocating workloads a fraction of a GPU rather than a whole GPU helped meet Exotel’s cost and performance requirements, Simplismart said.

NVIDIA collaboration and cloud deployment

Alongside NVIDIA’s investment, Simplismart said its engineers work directly with NVIDIA Solutions Architects, with technical contributions coordinated between the teams. It also said NVIDIA helps facilitate industry connections as Simplismart expands its enterprise and cloud-provider business.

Simplismart identified Yotta as a cloud partner, saying its software powers the serverless inference layer of Shakti Studio. In that arrangement, customers run models without managing the underlying infrastructure, according to Simplismart.

The company said deployment control remains part of its offering. Customers may need models to operate within a chosen cloud or on their own premises, rather than leave the location of computing and data to a service provider. Simplismart described data location and private-environment deployment as requirements its software is designed to accommodate.

Its proposed next layer addresses a different problem: matching changing demand with available computing capacity. Simplismart said an available GPU is not necessarily ready to answer requests because the required model may not yet be loaded. Keeping capacity ready for peak traffic, meanwhile, can leave customers paying for idle computing power when demand falls.

Simplismart said it is building towards scheduling workloads across regions and time zones. That remains a development objective: the available capacity must meet response-time requirements without moving workloads beyond a customer’s data-residency or compliance boundaries.