A new Seattle startup led by veteran engineers is building technology to tackle one of the biggest drivers of cost and complexity in large language model deployment.
Founded just a few months ago, ElastixAI is developing an AI inference platform designed to optimize how large language models are run.
The company just raised $16 million of a larger round led by Bellevue-based venture capital firm FUSE, GeekWire has learned.
- ElastixAI is led by CEO Mohammad Rastegari, who was co-founder and CTO at Xnor, a Seattle startup that specialized in edge-based AI tools and was acquired by Apple for around $200 million in 2020.
- ElastixAI co-founder Saman Naderiparizi, CTO, led hardware engineering at Xnor and was a senior engineering manager at Apple. The company’s third co-founder, Mahyar Najibi, was also at Apple and previously spent three years at Waymo.
As AI demand skyrockets, inference — how trained models “think” and produce answers — is one of the hottest areas of AI infrastructure.
ElastixAI is building a platform designed to improve the performance and efficiency of large language models — whether they are running on edge devices or larger cloud deployments, on a range of hardware configurations.
“We saw a gap when it comes to delivering AI inference at a scale and at low cost,” Rastegari told GeekWire. He described the platform as “mostly software-oriented.”
ElastixAI remains in stealth, but the founders say its platform could serve a range of customers — from hyperscalers to enterprises integrating AI into daily operations.
Part of the company’s secret sauce is providing flexibility.
“We want to enable customers to be able to actually configure their system to a very detailed level, so that they have an inference infrastructure that is built for their very specific use case,” Naderiparizi said.
The company’s other investors include Catapult, Tyche, Liquid 2 Ventures, and DNX Ventures.
“Thrilled to back the elite technical founders at Elastix to solve the hardest problems around scaling compute and infrastructure in the fast-growing AI inference market,” Cameron Borumand, general partner at Fuse, said via email.
There are several companies developing inference-related products, from larger players such as Nvidia and Coreweave to smaller startups including Fireworks.ai and Together.ai.
Naderiparizi said some of those companies could potentially be customers of ElastixAI.
Rastegari spent four years at Apple after the Xnor deal in 2020. He joined Meta last year as a distinguished scientist. He also spent five years at the Allen Institute for AI (AI2) as a research scientist.
Both Rastegari and Naderiparizi are affiliate assistant professors at the University of Washington.
Rastegari co-founded Xnor with Ali Farhadi, who spent three years at Apple before becoming CEO at Seattle-based AI2, created by the late Microsoft co-founder Paul Allen to boost AI research.
Apple has a growing Seattle-area footprint and earlier this year inked a big new lease in South Lake Union.
OctoAI, a Seattle startup that also helped companies run AI models more efficiently, was acquired by Nvidia last year.
Read the full article here
