Navyra launches Ale to cut repeated AI inference work

Navyra has launched Ale, a system designed to help businesses reuse previously computed AI inference work. The UK company says the product adds persistent memory to enterprise AI systems.

The launch comes as companies face growing pressure on GPU supply, data-centre capacity and electricity use as AI workloads rise. Navyra argues that many organisations waste computing resources because AI systems often repeat the same work when users or software agents ask identical or similar questions.

Ale sits in front of an organisation's existing inference engine and checks whether the work has already been done. If a request is an exact repeat and the output is deterministic, the system can return a result in under a millisecond. If a request is similar rather than identical, the underlying model checks the result before it is returned. New requests are processed in full by the model.

In tests, Navyra says organisations could handle roughly 1.6 to 3 times as many requests per GPU on repetitive workloads. That could reduce GPU time by about 37% to 67%, depending on how repetitive the traffic is.

Navyra says the product differs from semantic caching systems, which return responses judged to be close enough to earlier outputs. In testing across three model families, the company says 100% of verified reused answers were byte-identical to the output of the unmodified inference engine.

Energy pressure

The product enters a market where the cost and energy demands of AI infrastructure are under increasing scrutiny. Gartner forecast in June that global data-centre electricity consumption would reach 565 TWh in 2026, up 26% year on year, with AI-optimised servers accounting for 31% of data-centre power consumption.

Navyra argues that reducing repeated inference work could offer an alternative to simply expanding hardware fleets. Large organisations may be particularly exposed because employees and AI agents often request the same or similar information across fragmented corporate systems.

That could matter for companies trying to increase output from existing infrastructure without changing models or buying new hardware. Navyra says Ale works with GPUs already owned or rented by customers and can run in a customer's cloud, on premises or in air-gapped environments.

Dr James Schofield, Founder, Navyra, said: "The GPUs organisations already pay for are doing the same work again and again. Navyra gives the inference stack persistent memory, so work is computed once, verified and remembered, allowing the GPUs companies already own or rent to serve more traffic."

He added that the company's approach depends on verification rather than approximation. "What makes Navyra more than a cache is that nothing is ever reused on similarity alone. Exact repeats are replayed only where the model's output is deterministic, and anything merely similar is checked by the model before it is returned."

Assessment tool

Alongside the launch, Navyra has introduced a free profiler intended to help organisations estimate whether their workloads contain enough repetition to benefit from reuse. The tool runs inside the customer's environment and analyses existing logs to identify exact and semantically similar reuse opportunities, according to the company.

Navyra says the profiler is air-gapped and keeps data within the organisation. It is also designed to show when potential savings are limited, allowing customers to assess the likely impact before deploying Ale on live traffic.

Founded by computational biologist Dr Schofield, Navyra is based in Hythe, Southampton. It is targeting enterprise customers that run AI inference workloads and want to increase throughput without replacing their existing model stack.

Backing

Navyra has also secured investment from Oxford Technology, which is supporting the commercial rollout of Ale to its first enterprise customers and inference platforms. The investor has backed more than 200 science and technology start-ups since 1983.

The company has also been selected as one of the winners of Innovate UK's AI Champions: Frontier AI Discovery competition under its Fundamental AI theme. Innovate UK says the programme aims to identify new forms of AI development in support of UK-based companies.

Schofield said: "The AI industry has largely responded to growing demand by asking how we build more capacity. We think there is a more fundamental question: how much of that capacity is being unnecessarily wasted, repeating work that has already been done?"