#Quasa #QUA #Fractile
fragile is a UK-based deep tech company building specialized processors designed to radically accelerate frontier model inference. Their main innovation is to physically interleave memory and compute on the same silicon. This is an architecture that overcomes the traditional separation between memory and processing that currently limits both the speed and cost of large-scale language model inference.
By taking a full-stack approach (from transistor-level design to cloud inference software), Fractile aims to achieve both low latency and high throughput at the same time. The company claims its processors can run state-of-the-art models up to 25 times faster than existing solutions and at approximately one-tenth the cost, while efficiently delivering thousands of tokens per second to thousands of concurrent users.
This performance improvement is expected to not only optimize current deployments, but also unlock new possibilities. Significantly longer context windows and complex autonomous tasks (such as research or large-scale software development) that would take days to manually perform today could potentially be compressed into minutes.
Fractile has raised $220 million from investors including Accel, Factorial Funds, and Peter Thiel’s Founders Fund. The team operates in Bristol and London and combines deep expertise in hardware, software and AI systems.
Fractile is ideal for AI labs, cloud providers, and enterprises that require dramatically faster and more cost-effective inference on large-scale frontier models.
highlights
- innovative architecture Memory and compute are interleaved.
- Claims to be up to 25x faster inference At about 1/10th the cost.
- Designed for simultaneous low latency and high throughput.
- full stack development From silicon to software.
- strong funding and an experienced technical team.
-
Potential considerations
- Hardware still under development (Commercial chips are planned later).
- Current performance numbers Based on internal results and simulations.
- Most relevant to your organization Planning for large-scale inference infrastructure.
Overall rating: 4.6/5 stars
Fractile addresses one of the most critical bottlenecks in modern AI: the high cost and limited speed of frontier model inference. By redesigning the relationship between memory and compute at the silicon level, the company has the potential to meaningfully change the economics of running advanced AI models.
Review on Quasa.io and earn QUA rewards too!
Get started: https://quasa.io/projects/fractile
