Power & Sites · Europe

ARIA’s Scaling Inference Lab Deploys Lumai’s Optical Compute Tech

Credited to HPCwire · hpcwire.com · 2h ago

Lumai uses light to perform AI inference to lower data center power consumption by 90% and unlock 50x higher processing speeds

OXFORD, England, Nov. 3, 2026 — The Advanced Research and Invention Agency (ARIA), a UK public R&D funding agency, will deploy 3D optical inference technology from Lumai, an optical compute company building next-generation AI infrastructure.

With concerns about data center energy usage high on political and industrial agendas, ARIA will onboard Lumai Iris inference servers as the first optical inference platform in its Scaling Inference Lab – run by CommonAI CIC, a non-profit organization focused on collaborative AI development. The deployment will enable researchers to evaluate the value of optical compute in a commercial data center environment, as well as its potential to lower data center power consumption by 90%. Lumai Iris servers could also cut total cost of ownership to around one-tenth of conventional GPU architectures.

Lumai’s Iris technology runs production-scale large language models (LLMs), and works by carrying out AI computation within beams of light operating in three-dimensional space, rather than just using transistors in a silicon chip. Because numbers are encoded as light intensity and passive optical components, such as lenses, are used for mathematical operations, the computation itself uses a negligible amount of energy, which barely rises as performance scales.

“Power, not demand, now limits how fast AI can be deployed.” said Dr. Xianxin Guo, co-founder and CEO of Lumai. “Running a multi-node system in a commercial data center is where new compute architectures prove whether they are ready for real deployment. This collaboration at ARIA’s Scaling Inference Lab validates our roadmap with a real system operating in a commercial environment.”

ARIA is an independent, UK government-backed funding organization supporting advanced AI and other transformative technologies. Created by an Act of Parliament, and sponsored by the Department for Business, Innovation, Science and Trade, it champions scientists and engineers pursuing ambitious ideas that could lead to major scientific and technological breakthroughs. Particularly those that benefit wider society. Its Scaling Inference Lab exists to stress-test next-generation compute architectures on real workloads in a live commercial setting – using standard power, cooling, and rack integration.

“Lumai’s technology is the most mature optical compute platform we evaluated, and has the potential to make AI inference more energy-efficient and less dependent on constrained global hardware supply chains,” said David Lewis, ARIA’s Chief Science and Technology Officer. “The Scaling Inference Lab provides the means to test that potential openly and independently at rack scale.”

Creating data center efficiencies will be essential as the demand for AI inference accelerates. Traditional AI inference runs electronically in silicon, which is power-hungry at scale. If relying only on these conventional accelerators to meet the computation ambition of even just one leading AI lab, it is estimated that the infrastructure bill would reach $100 trillion within the next five years. Added to this, the additional energy requirement to be in excess of 1,000 GW. These energy and cost requirements would make it impossible to meet this ambition without a change in the physics of AI computation.

For the UK, the challenge is also one of sovereignty. Building the domestic AI compute capacity the country needs will depend on getting more from limited power and grid capacity, and on bringing UK-developed technologies into real data centers.

“As the UK builds out its sovereign AI compute requirements, it has to do so within the power it has. We need to find and then prove new, innovative approaches to AI compute,” said Gavin Ferris, CEO of CommonAI. “The Scaling Inference Lab was set up for exactly that – to test innovative technology, such as Lumai’s optical compute, on real workloads alongside today’s systems.”

Tackling Compute-Heavy Prefill Workloads

Large-model inference happens in two phases: prefill, which is compute-bound, and decode, which is memory-bandwidth-bound. While specialized inference hardware has increasingly focused on decode, Lumai’s approach is different. It targets the compute and matrix-heavy prefill stage before the first token is generated. Lumai Iris performs millions of matrix operations in a single optical cycle.

Rather than a replacement for silicon, Iris servers are particularly effective for prefill when paired with silicon for decode – delivering at least 16x better energy efficiency per inference than silicon-only systems, addressing the power and cost ceilings now constraining AI deployment.

The ARIA deployment builds on the April 2026 launch of Iris Nova, Lumai’s first product available for evaluation and the first optical AI accelerator engineered for data center-scale inference. ‍

Models & LabsEurope

Britain on Competing in the AI Race

Credited to Bloomberg$ · bloomberg.com1h ago

Useful

Kanishka Narayan, Minister of State for Artificial Intelligence, United Kingdom, discusses how Britain can turn its strengths in AI research into global companies, higher productivity and strategic advantage with Bloomberg’s Tom Mackenzie at Bloomberg Tech 2026 in London.…

View the article · $Bloomberg

Front

Models & Labs · Europe

British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted

Credited to The Register · theregister.com · 2h ago

Photograph · The Register

Revolut says that in less than a year it's managed to slash its AI coding inference costs by roughly 90 percent by making self-hosted models its default instead of relying primarily on commercial services like Claude.

Speaking at the AI Engineer conference in New York last month, David Abitbol, ops and infrastructure lead at the British fintech, described how it saw a 60 percent improvement in coding time and a 70 percent improvement in lead time following its rollout of AI coding tools to roughly 2,000 engineers. Revolut initially adopted GLM-5.1 before moving to GLM-5.3, which it now serves using 56 Nvidia B200 GPUs.

Prior to its switch to GLM, Revolut had been experimenting with AI, deploying Gemini in early 2025 and Claude Code in early 2026. "We were highly skeptical about the cost of AI. We were seeing engineers spend $1,000 a month, $10,000 a month with not so much ROI, and so when we launched it we launched it with strict budget limits," Abitbol said.

Because of this, engineers only had access to a limited number of tokens, so the company began investigating self-hosting its own models. At $0.30 per million input/output tokens, self hosting GLM 5.1 cost about a tenth of using OpenAI's Codex or Anthropic's Claude Code. And, as it turned out, when tokens are cheap, you stop worrying about how many they use or what they use them for.

While Revolut did see improvements to coding speed, Abitbol noted the biggest bottleneck wasn't actually the code. Instead, he says the biggest gains came from letting the AI models deal with things like debugging and Jira and Confluence tickets. "When you have unlimited tokens, you're free to do that," he said. "You no longer have to save your tokens for the end of the quarter or the most difficult task."

Developed by Chinese open weights model dev Z.ai, GLM 5.1 isn't exactly a small model. At roughly 754 billion total parameters (40 billion activated for each token generated), the model requires roughly 380 GB of GPU memory just to hold the weights and another 6-26 GB per user to maintain context depending on precision.

To serve GLM-5.3 at scale, Revolut deployed 56 Nvidia B200 GPUs, equivalent in GPU count to seven eight-GPU DGX B200 systems. But as Abitbol noted, the greatest performance gains didn't come from adding GPUs, but rather tuning the model's hyperparameters along with switching to NVFP4 quantization, implementing disaggregated prefill and decode, and tuning context lengths.

One key takeaway was that supporting a one-million-token context window doesn't necessarily make such a large window useful or efficient. "Long context is a scam. You really need to be mindful of that," Abitbol said. "Even though you can serve one million tokens, 256 [thousand] is really what you need."

With these optimizations, Abitbol says that the company was able to achieve sub 3-second time to first token on context windows of around 256,000 tokens, and generation speeds of around 200 tokens a second, both of which he notes are the sweet spot for the company's software engineers.

"What you want to aim for is a sub-3 second time to first token. This is empirical evidence. As soon as you exceed 5 seconds, my Slack inbox starts to blow up," Abitbol said. "In terms of throughput, it doesn't really matter as much. We serve, in terms of interactivity, 200 tokens per second. People are fine [with that]. You don't need to go higher. The way people work has changed. We no longer look at the stream of thinking tokens. People run tasks in parallel and check in when the task is done."

And if throughput is a concern, he notes engineers who sign in after 5 pm are getting closer to 800 tokens a second.

However, it's worth noting that GLM hasn't replaced GPT or Claude entirely. The Chinese model is text only, so it isn't ideal for things like UI work. So instead, Revolut used a model router to direct vision queries to models like Gemini or GPT Luna while keeping code generation running on GLM.

"Open source is now a priority at Revolut, and our economics are favorable. The billing is flat, we don't have any bill shock at the end of the month, and the frontier is the exception," Abitbol said.

Having said that, Abitbol acknowledges that Chinese open weights models may be a concern for privacy and security reasons, but emphasizes that the major concern there is accessing them via APIs running out of China. When run on-prem or on regional cloud resources, there's really nothing to worry about. ®