What ElastixAI does

ElastixAI is a Seattle-based AI infrastructure startup building a software-ML-hardware co-design platform for running large language model inference on FPGA-based servers. Founded by former Apple and Meta machine learning researchers, the company targets hyperscalers, AI model providers, and large enterprise AI deployments that face rising cost and energy constraints from GPU-centric inference infrastructure.

The platform applies proprietary post-training ML optimizations — including extreme quantization (4-bit and binary), sparsity, and mixture-of-experts routing — and automatically generates a custom processor design optimized for each specific model, deploying the resulting bitstream to an off-the-shelf FPGA in seconds. This creates what the company terms an "ML-defined, software-configurable compute" architecture, delivering a claimed 5-50x total cost-of-ownership advantage per token and roughly 80% lower power consumption versus GPU-based inference. Distribution is engineered for frictionless adoption: the system ships as a drop-in backend replacement for the NVIDIA CUDA plugin and OpenAI API stack, with native PyTorch and vLLM support and no application code changes required.

ElastixAI pursues a direct enterprise sales motion and monetizes primarily through software licensing of its FPGA inference platform, with potential associated revenue from FPGA hardware deployments. The company emerged from stealth in February 2026 alongside an $18 million seed round led by FUSE and joined by Catapult Ventures, DNX Ventures, Liquid 2 Ventures, and Tyche Partners (with the founder network of former Apple and Meta ML researchers also participating). The leadership team comprises five named executives — co-founders Mohammad Rastegari (CEO), Saman Naderiparizi (CTO), and Mahyar Najibi (CSO), plus James Allard (CFO/COO) and Jon Gelsey (Head of Strategy and Marketing) — with total headcount in the 11-50 range.

Name

ElastixAI

Legal name

ElastixAI Inc.

Website

https://elastix.ai

Company type

Private

Founded year

2026

Operating status

Operating

Headcount range

11–50 employees

Short description

Seattle-based AI infrastructure startup founded by former Apple and Meta ML researchers, building a software platform that converts off-the-shelf FPGA servers into AI supercomputers for generative AI inference, delivering a claimed 5-50x TCO-per-token advantage over GPU backends.

Ownership category

akta.pro rank

Product category

AI Inference Infrastructure

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182), Computer Systems Design and Related Services (5415), Computer Systems Design and Related Services (54151)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC)

akta.pro secondary industries

Model Hosting, Serving & Inference Platforms (HDAAACAB), Model Deployment, Serving & Inference Platforms (HDAAABAF), AI Server Systems & HGX/Accelerator Platforms (HDAAAAAB)

Keywords

• AI inference platform
• FPGA acceleration
• LLM inference optimization
• Hardware-software co-design
• AI infrastructure software

Headquarters

HQ city

Seattle

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Marketing or Sales, Operations, Infrastructure

Revenue model

1. FPGA Inference Platform Software: Software platform that delivers efficiency gains through hardware-software co-optimization for LLM operations, sold as a replacement for legacy GPU workflows. The platform enables cost savings through lower CapEx (FPGAs significantly cheaper than high-end B200 or H100 cards) and lower OpEx (higher compute utilization and 80% lower power consumption). Revenue is derived from software licensing and potentially hardware component sales associated with the FPGA deployment.

Go-to-market motion1 record

Distribution channels1 record

Marketing channels4 records

Core offering

ElastixAI develops a software-ML-hardware co-design platform that converts off-the-shelf FPGA-based servers into AI supercomputers for generative AI inference. The platform applies proprietary post-training optimizations to a user-provided LLM, automatically generates a custom processor design tuned to the model's algorithms and bit widths, and deploys it to an FPGA in seconds as a drop-in replacement for NVIDIA GPU backends. It maintains compatibility with existing OpenAI API, PyTorch, and vLLM workflows with no code changes required.

Product overview

ElastixAI offers a unified platform consisting of the core ElastixAI Platform that provides ML-defined, software-configurable FPGA compute, supplemented by the Elastix Plugin and Elastix Shell/API which together function as a drop-in replacement for NVIDIA GPU backends. The platform enables users to convert off-the-shelf FPGA-based servers into AI supercomputers while maintaining compatibility with existing OpenAI API, PyTorch, and vLLM workflows.

Differentiator

Problem solved

Functional benefit

Products and services

• ElastixAI FPGA Inference Platform ML-defined, software-configurable compute platform that converts off-the-shelf FPGA-based servers into high-efficiency AI supercomputers for generative AI inference. The platform applies proprietary post-training optimizations to a user-provided LLM, automatically generates a custom processor design optimized for the model's algorithms and bit widths, and deploys it to an FPGA in seconds as a drop-in replacement for NVIDIA GPU backends. Maintains compatibility with existing OpenAI API, PyTorch, and vLLM workflows with no code changes required. Targets hyperscalers, AI model providers, and enterprise AI deployments.

Quantifiable outcome

• 5-50x TCO per token advantage over GPU-based solutions
• +4 more outcomes

Segments2 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration3 records

AI capability3 records

Feature5 records

Scale indicators6 records

Recent moves7 records

Expansion highlights5 records

Broad incumbents

• AMD (Xilinx): Primary supplier of the off-the-shelf FPGAs (Versal, Alveo) that ElastixAI's platform targets, and itself a competitor in adaptive compute for AI inference. Comparable both as an enabling vendor (FPGA supply) and as a broad incumbent in reconfigurable AI compute.
• NVIDIA: The dominant AI accelerator supplier whose CUDA-based GPU stack (H100, B200) is the incumbent ElastixAI explicitly targets as a drop-in backend replacement. Comparable because both serve hyperscalers and enterprise inference workloads, but NVIDIA is a broad-platform incumbent while ElastixAI focuses narrowly on reconfigurable inference.

Direct peers

• Groq: Specialty AI inference silicon company offering a custom LPU architecture that competes head-on with ElastixAI for hyperscaler and enterprise inference customers. Both companies pitch dramatically lower latency and cost-per-token than NVIDIA GPUs as their primary differentiation.
• Tenstorrent: AI accelerator company developing RISC-V-based Grayskull/Wormhole/Blackhole chips for training and inference. Comparable as an alternative-silicon vendor with proprietary hardware-software co-design and a focus on competing with NVIDIA for hyperscaler workloads.
• SambaNova Systems: AI accelerator company building RDU-based systems and a full software stack (SambaSuite) for enterprise and hyperscaler LLM inference. Directly comparable in product category, customer profile, and the pitch of dramatically better cost-per-token than GPUs.
• Lightmatter: Photonics-based AI inference and interconnect company pitching dramatic energy efficiency gains for LLM serving. Comparable because it sells alternative AI inference compute to hyperscalers with a similar energy/cost-per-token value proposition.
• Cerebras Systems: AI accelerator company using wafer-scale chips to deliver high-throughput inference and training. Compares to ElastixAI as another alternative-AI-silicon vendor targeting the same hyperscaler buyers seeking to escape NVIDIA GPU constraints.
• Graphcore: AI accelerator company (IPU) competing in inference and training for enterprise and cloud customers. Directly comparable as a non-GPU AI silicon vendor targeting the same TCO-per-token economics that ElastixAI emphasizes.

Emerging players

• Positron: AI inference accelerator company focused on memory-bandwidth-bound transformer workloads, recently rebranded from NeuroBlade. Comparable as a smaller emerging competitor targeting the same inference TCO-per-token problem ElastixAI addresses.
• Lattice Semiconductor: Low-power FPGA vendor (Avant, Certus) increasingly targeting AI inference at the edge and in data centers. Comparable as the second major supplier of FPGAs ElastixAI-compatible platforms depend on, and as an emerging player in AI-accelerated FPGA silicon.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat4 records

Key risks6 records

Key highlights6 records

Customer concentration

• LinkedInlinkedin.com/company/elastixai

Revenue estimate

Valuation estimate

Number of profiles

Profiles5 records

Funding overview

Funding rounds3 records

Investors7 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about ElastixAI