In an October 27 letter, CEO Amritanshu Jain announced Simplismart's $14 million Series B, led by Dallas Venture Capital, with NVIDIA participating.

The big change

  • What changed: The round funds software for configuring and operating inference services.
  • Why it matters: Buyers can compare usage-priced APIs, dedicated GPUs and private deployment against the way their own applications consume capacity.
  • What to watch: Jain's proposed cross-region scheduler remains a future build, with latency and data-location limits to satisfy.

What the current controls expose

The Simplismart SDK reference for version 0.1.14, released October 7, documents a dedicated deployment built from a model repository, GPU selection, replica bounds and a GPU-use scaling target. It includes a deployment health-check method and a separate bring-your-own-cloud request naming the customer's node group.

The autoscaling reference also shows a p95 latency target and scale-to-zero with a cooldown. Its scale-to-zero example uses a minimum replica setting of one. That setting does not establish actual capacity or a response-time guarantee. BIG CHANGE did not deploy or benchmark this service.

Three ways to buy capacity

Simplismart's pricing page separates usage-based model APIs from dedicated deployments priced by GPU hour. It offers private-cloud and on-premises deployment through consultation, without a public standard rate for those arrangements. These are different billing bases. Compare per-token quotes only after aligning the model and input/output token billing; a GPU-hour quote also needs expected running hours and useful output to estimate unit cost.

For a purchasing comparison, record the model, expected request volume over time and the bill under each eligible arrangement. The public price list does not settle private-deployment terms. Those inputs frame a buyer's cost evaluation; they are not a measured Simplismart saving.

One customer example, with a narrow claim

Jain reports that Simplismart met Ema's approximate 200-millisecond and nine-cent-per-million-token targets at 180 milliseconds and seven cents per million tokens. The letter gives no test protocol or independently verified result.

These controls support a buyer's evaluation; the documentation does not establish performance or answer quality on that workload.

The roadmap

Jain says Simplismart is building toward matching demand with third-party GPU capacity across regions, subject to latency, residency and compliance limits. The letter gives no availability date or public test for that proposed global scheduler.