What ElastixAI Builds
ElastixAI builds a reconfigurable AI inference stack that co-designs machine learning optimizations, serving software, and FPGA-based processors into a unified execution layer. On its official website, the company frames the hardware bottleneck directly: model-level efficiency techniques arrive continuously, while fixed custom ASICs take years to tape out. ElastixAI automates FPGA reconfiguration so a single processor architecture can adapt across dense language models, mixture-of-experts (MoE) architectures, and newer model topologies without forcing enterprise teams to write low-level hardware code.
The platform targets two deployment paths: hosted LLM inference through an OpenAI-compatible API and on-premise installations delivered through a PyTorch-compatible flow meant to swap into existing GPU environments. On elastix.ai, ElastixAI argues that modern FPGAs offer favorable cost per gigabyte of attached memory and memory bandwidth—the primary economic constraints on large-model token generation—and publishes self-reported efficiency targets of 10x+ more tokens per dollar, up to 80% lower power draw, and 5–10x lower energy per token compared with conventional GPU deployments.
- Live hosted model catalog: Lists 4-bit hosted inference pricing on elastix.ai for OpenAI
gpt-oss-20b($0.04 input / $0.14 output per 1M tokens, 131K context),gpt-oss-120b($0.08 / $0.33, 131K context),Qwen3.8 27B($0.07 / $1.84, 262K context), andQwen3.6 35B A3B($0.07 / $0.88, 262K context). - Upcoming model roadmap: Displays coming-soon hosted endpoints for Xiaomi
MiMo-V2.6-Flash(1M context), TencentHy3(262K context), and Z.aiGLM 5.3andGLM 5.3 Flash(262K context). - Zero-FPGA-programming integration: Exposes OpenAI-compatible API endpoints for hosted workloads and a PyTorch-compatible on-premise workflow intended to automate hardware specialization transparently.

| Field | Verified Record (As of Nov 19, 2026) |
|---|---|
| Company | ElastixAI (ElastixAI Inc.) |
| Official Website | https://www.elastix.ai/ |
| Washington HQ | Seattle, WA (1505 Westlake Ave N #320, Seattle, WA 98109) |
| Other WA Footprint | Plug and Play Seattle Batch 3 (access to UW CoMotion Labs office, Spring 2026); co-founders hold UW affiliate assistant professor appointments |
| Founded | Early 2025 (GeekWire, May 14, 2025 & Apr 16, 2026) |
| Founders / Leadership | Mohammad Rastegari (Co-founder & CEO), Saman Naderiparizi (Co-founder & CTO), Mahyar Najibi (Co-founder & CSO) |
| Stage | Seed |
| Announced Funding | Single May 14, 2025 Seed round led by FUSE ($16M reported by GeekWire; $18M reported in company announcement), with Catapult Ventures, Tyche Partners, Liquid 2 Ventures, and DNX Ventures |
| Team Size | Not publicly disclosed |
| Primary Ecosystem Ties | FUSE (Bellevue VC); Xnor.ai & AI2 founder lineage; University of Washington; Plug and Play Seattle Batch 3; 2026 GeekWire Awards Startup of the Year finalist |
Washington Footprint and Leadership
ElastixAI operates from Seattle, listing a corporate address at 1505 Westlake Ave N #320, Seattle, WA 98109 on elastix.ai. The founding team carries deep ties to Seattle's computer vision, edge-AI, and academic research communities. Co-founder and CEO Mohammad Rastegari previously co-founded and served as CTO of Seattle AI2 spinout Xnor.ai, which Apple acquired for roughly $200M in 2020; he spent four years at Apple, five years as a research scientist at the Allen Institute for AI (AI2), and later joined Meta as a distinguished scientist (GeekWire, May 14, 2025).
Co-founder and CTO Saman Naderiparizi led hardware engineering at Xnor.ai before becoming a senior engineering manager at Apple, and both Rastegari and Naderiparizi hold affiliate assistant professor appointments at the University of Washington (GeekWire, May 14, 2025). Co-founder and Chief Science Officer Mahyar Najibi joined from Apple after previously spending three years at Waymo. ElastixAI has not publicly disclosed its current employee headcount in company materials or GeekWire reporting.

Products, Users, and Public Traction
ElastixAI targets enterprises, AI application developers, and data center operators running high-volume language model inference where power limits and token costs constrain deployment. In addition to its hosted API catalog and interactive homepage demo, elastix.ai maintains an active technical blog updated through November 9, 2026 covering quantization, memory-and-storage bottlenecks, and tokens-per-second-per-watt economics. Its homepage features infrastructure and hardware collaborations with Steiger Dynamics (CEO Martin Gossner) and Seattle-based Sabey Data Centers (Sabey Corporation CEO John Sabey).
Paying enterprise customer counts and annual revenue are not publicly disclosed. Public ecosystem validation in Washington includes ElastixAI's debut on the Q2 2025 GeekWire 200 (entering at No. 187 shortly after stealth emergence), selection for Plug and Play Seattle's third accelerator cohort in March 2026 (with workspace access at UW CoMotion Labs), and selection as a Startup of the Year finalist at the 2026 GeekWire Awards.
Funding, Stage, and Place in Washington's AI Landscape
ElastixAI is a Seed-stage company with a single publicly disclosed institutional financing announced on May 14, 2025, led by Bellevue-based FUSE with participation from Catapult Ventures, Tyche Partners, Liquid 2 Ventures, and DNX Ventures. Public sources differ slightly on the final close size of that single round: GeekWire reported $16M "of a larger round," while the company's May 14, 2025 emergence announcement stated $18M. No subsequent funding round or acquisition has been announced through November 2026, and ElastixAI remains independently operated at elastix.ai.
Within Washington's AI ecosystem, ElastixAI represents Seattle's enduring depth in systems-level AI efficiency—following local predecessors such as Xnor.ai (acquired by Apple) and OctoAI (acquired by Nvidia in 2024). By pairing reconfigurable FPGA execution with standard OpenAI and PyTorch software interfaces, the company competes in the infrastructure tier alongside GPU cloud providers and model-serving startups while addressing rack power constraints highlighted across Pacific Northwest data center deployments.
Related Washington AI Profiles & Tygart Media Context
- Ascerta — Bellevue AI Profile
- StitcherAI — Seattle AI Profile
- Loopr AI — Bellevue AI Profile
- Why AI Compute Needs a Warranty, Not Another Watermark
Verified Primary Sources
- ElastixAI Official Website
- GeekWire — ElastixAI $16M Seed Round (May 14, 2025)
- GeekWire — Plug and Play Seattle Batch 3 Cohort (Mar 23, 2026)
- GeekWire — 2026 GeekWire Awards Startup of the Year Finalists (Apr 16, 2026)
- GeekWire — Q2 2025 GeekWire 200 Reboot (Jun 26, 2025)
Get The Signal
What we actually ran, what it measured, and what it means for a restoration shop — every week, including the flops.

Leave a Reply