ElastixAI Emerges From Stealth to Redefine Generative AI Economics via FPGA-Based Supercomputers

SEATTLE--( BUSINESS WIRE)-- ElastixAI Inc. today emerged from stealth to tackle the systemic inefficiencies and high costs of generative AI (GenAI) inference. Founded by former Apple and Meta machine learning (ML) researchers, and backed by $18M in seed funding, the company is launching a novel software platform that converts off-the-shelf FPGA-based servers into high-efficiency AI supercomputers.

“We’re moving away from 'one-size-fits-all' hardware. By applying proprietary post-training optimizations to FPGAs, we let hardware adapt to the model rather than forcing the model to struggle on the hardware," said Mohammad Rastegari.

The AI inference market is expected to reach $255 billion by 2030, yet existing infrastructure is fundamentally mismatched for GenAI. While LLM inference is memory-bound, standard GPUs are designed for compute-bound tasks like training. This results in very low compute utilization during inference, leading to massive wasted capital and energy. Further, custom silicon is architected years before it reaches production, leaving it years behind current ML innovations. For example, 4-bit quantization—which theoretically doubles performance—often yields only a 10% gain on hardware lacking native support.

“The industry is currently leaving an order-of-magnitude of performance on the table because hardware can’t keep up with the advances in ML,” said Mohammad Rastegari, PhD, co-founder of ElastixAI. “We’re moving away from 'one-size-fits-all' hardware. By applying proprietary post-training optimizations to FPGAs, we let hardware adapt to the model rather than forcing the model to struggle on the hardware."

ElastixAI provides a drop-in replacement for legacy GPU workflows, maintaining workflow compatibility while delivering massive efficiency gains. Key benefits include:

ElastixAI is available to select enterprise partners, data center operators, and AI model providers. Visit www.elastix.ai for technical specifications or to request a demo.

About ElastixAI Inc.

ElastixAI solves the systemic inefficiencies of GenAI inference through innovative software-ML-hardware co-design, delivering the next generation of scalable, sustainable AI. The founding team brings unparalleled expertise in ML, SW, and HW from the world’s leading technology companies and academic institutions, with a combined 24,000 citations.