Skip to content

MachGen AI

4 viewers · 30d

Total raised

$10.0M

+$10.0M this year

1 filing since 2026 · latest Equity filed

Cumulative raised
LAST ROUND
Equity · $10.0M
ROUNDS
1
INVESTORS
3
FOUNDED
2025
HQ
Milpitas, CA
SECTOR
AI
EMPLOYEES
N/A
30D VIEWERS
4

AI overview

Updated

MachGen AI builds a high-performance inference and fine-tuning stack purpose-built for diffusion models, covering image, video, and emerging world models. The company optimizes open-source generation models at the kernel level - attention, caching, kernels, and parallelism - so they run at a fraction of the usual latency and cost while producing the same output from the same weights. Published benchmarks claim roughly 4-6x faster image generation and about 6x faster video generation, at 2-4x lower cost than incumbent providers. It was founded in 2025 by Kismat Singh and Manoj Krishnan, engineers who previously worked on TensorRT inference at Nvidia, datacenter AI software at Intel, vLLM at Google, and PyTorch training infrastructure at Meta.

What sets it apart

A diffusion-specific inference stack rather than an LLM stack retrofitted for images and video - kernel-level optimization that keeps outputs bit-for-bit equivalent to the original weights while cutting latency several-fold, which independent benchmarking placed in the fastest and lowest-cost quadrant.

Funding history

1 round
EquityFeb 10, 2026
Form D/A
+$10.0M$10.0M total

Latest SEC filings

via EDGAR · CIK 0002109967
Form D/A · Feb 10, 2026View on EDGAR

Products

1 tracked

MachGen inference platform

AI inference infrastructure

An inference platform for diffusion and world models, offered as optimized model APIs across popular open models (Flux, HiDream, LTX, Wan, MiniMax and others) plus a managed cloud service.

  • Kernel-level optimization of the full inference stack (attention, caching, kernels, parallelism)
  • 4-6x lower latency on image models and roughly 6x on video generation
  • 2-4x lower inference cost versus other providers
  • Model APIs for popular open diffusion models