What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Lumai is building optical hardware for a specific part of AI inference: processing a model’s input context before it generates tokens. Its approach sends light through a three-dimensional free-space volume to perform matrix operations in parallel, rather than relying only on computations confined to electronic chips. Lumai says this could improve performance and energy use, but the available public material does not independently verify those comparative claims. The company’s Iris Nova server is described as built, validated on Llama 3, and available for evaluation.
What Lumai means by free-space optical computing
In Lumai’s account, its approach grew out of Oxford research into using light to perform matrix operations—the repeated calculations central to many machine-learning workloads. Instead of confining computation to electronic circuits on a flat chip, beams of light propagate through a three-dimensional optical volume. Lumai says that spatial arrangement lets many operations happen in parallel in the optical domain.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Nimo AI NAS, Agentic Computer Mini PC and AI Server, AMD Ryzen 7 PRO 8845HS(up to 5.1 GHZ, beat... | $1,999.99 | Buy on Amazon |
This is different from using fiber-optic cables simply to move data between conventional processors. Lumai’s proposal is to use light for computation itself, then combine that optical compute with conventional hardware in an AI server. The company frames spatial parallelism as a route to faster, lower-energy matrix operations; the public sources do not provide independent measurements that quantify that benefit.
Why Lumai is targeting inference prefill
LLM inference has two broad stages. Prefill processes the prompt and its context; decode generates the response one token at a time. Lumai describes prefill as compute-bound and decode as memory-bound, and says its optical hardware is aimed first at prefill.
#1 Best Overall
- [Local AI Inference & 70B Model Ready] Equipped with the AMD Ryzen 7 PRO 8845HS processor, NEXUS is engineered for heavy local AI workloads. With a full-size GPU bay, it runs 70B LLMs natively without an internet connection. Ideal for AI developers and tech enthusiasts who need private environment for coding and model testing.
- [132TB Mass Storage with ZFS Integrity] Features a hybrid storage architecture (3×NVMe + 4×3.5" HDD) supporting up to 132TB. Utilizing the enterprise-grade ZFS file system and ECC memory, it prevents data corruption and bit rot—a must-have for professional photographers and video editors safeguarding 4K/8K RAW footage.
- [OpenClaw-Driven Automation Workflow] The built-in OpenClaw execution layer allows complex automated tasks to be processed locally. Even when offline, your backup schedules and AI file organization continue seamlessly. Say goodbye to monthly cloud subscriptions and high latency.
- [Dual 10GbE & USB4 Ultra-Connectivity] Experience server-class speeds with dual 10GbE ports and a 40Gbps USB4 interface. It enables multi-user real-time collaboration on large project files directly from the NAS, ensuring zero-lag editing for creative studios and production teams.
- [Open-Source ZimaOS for Total Privacy] Running on the fully open-source ZimaOS, NEXUS ensures your data stays physically on-premise with no backdoors. It acts as a "Digital Fortress" for privacy-conscious families and small businesses who demand absolute data sovereignty.
The company’s proposed design is disaggregated: use optical compute for prefill and conventional hardware for decode. That is a workload strategy, not evidence that every model, context length, or serving setup will benefit. A meaningful comparison would need to hold those conditions constant and measure the complete system, not just an optical calculation.
What Lumai says about Iris Nova and its roadmap
Lumai calls Iris Nova its first-generation optical AI server. In a September 15, 2026 announcement, the company said it runs billion-parameter models, had been validated on Llama 3, and was available for evaluation. Lumai CEO Xianxin Guo described it as “real hardware, ready for evaluation today.” The announcement also says it can be deployed in existing air-cooled data-center racks. These are company-reported product-status claims; they do not establish broad customer deployment or independently verified production readiness.
| System | Lumai’s stated position | Status and qualification |
|---|---|---|
| Iris Nova | First-generation optical AI server using discrete photonic components | Lumai said in September 2026 that it was built, validated on Llama 3, and available for evaluation. |
| Iris Aura | Intended to move toward integrated photonic devices | Lumai’s June 2026 roadmap gave an approximate two-year target horizon. This is forward-looking company guidance, not a confirmed delivery date. |
| Iris Tetra | A later, more comprehensive solution | Described in Lumai’s June 2026 roadmap; timing and final scope were not stated. |
The progression from discrete components toward integration makes manufacturing a central question for the roadmap, rather than a problem the first server has already resolved.
Recommended Free Tools
Why packaging and manufacturing are part of the technical challenge
An optical computation is only one part of a deployable system. Components must be aligned, connected, packaged, integrated with the rest of the server, and made to work within data-center operations. Lumai’s June 2026 article identifies fiber attachment as a particular scale-up challenge: the company says active-alignment attachment can take two to three orders of magnitude longer than wirebonding or flip-chip bonding. That is Lumai’s comparison, not an independently established industry measurement.
Lumai points to automated high-density fiber attachment, passive alignment, and standardized connectors as areas of work. These details matter because manufacturing time, repeatability, repair, and system integration can affect whether a promising optical device can be produced and operated at useful scale. The public material does not establish optical packaging yield, long-term stability, or serviceability for Iris Nova.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What performance claims can—and cannot—show
Lumai CEO Xianxin Guo said in September 2026 that the system could use roughly 10 times less energy per inference than GPUs. The company’s homepage also states 50 times performance and 90% lower power, but does not provide enough detail there to establish a like-for-like comparison. These claims should not be combined: they use different measures, and the public material does not give independent benchmark methods for either.
To judge the claims, evaluators would need results on defined workloads and comparable systems. Relevant details include model and context size, throughput and latency, energy per request or token including conversion and host hardware, total system cost, rack and cooling needs, and software integration. No independent comparative benchmark or deployment evidence establishing Iris Nova’s performance, energy use, cost, or reliability was identified in the cited public material.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →In an April 28, 2026 Unite.AI interview, Guo summarized the challenge as moving beyond a laboratory demonstration: “The challenge was never demonstrating that optics could perform computation – researchers had shown that in principle for years. The challenge was making it work at scale, outside the lab.” That distinction is useful: demonstrating an optical operation is not the same as showing a reliable, economical, end-to-end inference server.
Quick Recap
What to watch in an evaluation
- Workload fit: whether the benefit appears in prefill for the tested models and context sizes, rather than being inferred from a different task.
- End-to-end measurements: latency, throughput, and energy for the complete serving system, including conventional decode hardware and optical-to-electronic conversion.
- Operational evidence: repeatability, uptime, serviceability, and the ability to fit the system into real data-center workflows.
- Scale-up: manufacturing yield and whether packaging and fiber-attachment methods can support repeatable production.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
