Escape Meridian

Dina Genkina · Nov 17, 2026

IEEE Spectrum: AI Is Giving Optical Computing Another Shot

Markdown · JSON · Original source

’s Dina Genkina describes how pressure to reduce AI inference’s energy use is renewing interest in optical computing: using light to perform calculations. The appeal is that matrix-vector multiplication, a calculation repeated in AI inference, can in principle use less energy with light because the multiplication itself can be passive. But encoding data into light and converting results back to digital form also consume energy, and inference includes operations that optical systems may not handle as well. The article follows several approaches: free-space optics, integrated photonics, and adding calculations to data-center communication paths. Their targets differ, from inference prefill—the phase that processes an input—to decode, which is more constrained by moving data between memory and processors. Company performance figures remain claims or, in one case, results from emulation; they do not establish energy savings in deployed systems. Genkina’s account also cautions against assuming optical computers will replace : the technology may instead complement them or support different computations. The central tension is whether optical systems can deliver useful efficiency at scale despite conversion costs, manufacturing constraints and the portions of workloads they cannot accelerate.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

Close-up of a green circuit board with a central microchip and intricate traces
IEEE Spectrum · Close-up of a green circuit board with a central microchip and intricate traces · Open original source

Original source

· Article

Source details

Original source date: Nov 17, 2026 (2026-11-17T14:00:03+00:00).

Date basis: Date stated by the publisher.

Added to Escape Meridian: 2026-11-18T11:34:50.212Z. First observed: 2026-11-17T14:06:48.267Z. Retrieved: 2026-11-17T19:16:06.011Z.

Referenced by

Recorded citations in public readings. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

The article reports company targets and claims, not demonstrated system-wide energy savings at scale. It says the energy advantage of integrated photonics is “yet to be conclusively demonstrated.” LightSolver’s reported speed results are from emulation, and the team is still working on explicitly energy improvements. The article describes potential uses across several optical-computing approaches; it does not establish which approach, if any, will prevail.

Open question

Can optical-computing companies deliver their targeted 10- to 80-fold energy-efficiency improvement at scale before traditional electronics catches up?

Source-stated

Dina Genkina reports that AI inference’s energy demands have renewed interest in optical computing as a possible way to reduce power use.

AI inference—the process of using AI models to respond to user queries—has driven much of the construction of huge, power-hungry data centers

Source-stated

The article explains the proposed energy advantage: optical multiplication can be passive, but encoding inputs and reading results also cost energy, so system size and conversion costs matter.

The only cost is encoding the vector into light and reading out the answer

Source-stated

Optical systems target different parts of AI workloads: some focus on prefill, while Optalysys aims to add computation to communication paths and memory-bound decode.

Lumai, and other companies in the space, are focusing on the compute-intensive prefill phase of inference.

Source-stated

The article reports that optical-computing companies are aiming for a 10- to 80-fold improvement in energy efficiency over today’s GPUs, while asking whether they can achieve it at scale before electronics catches up.

optical-computing companies today are aiming for a 10- to 80-fold improvement in energy efficiency over today’s GPUs.

Source-stated

Genkina presents optical computing as a possible complement to GPUs rather than an assured replacement, with different hardware potentially suited to different tasks.

We see the data center of the future as heterogeneous: CPUs, GPUs and photonic NPUs [native processing units] each take on the tasks they are best suited for

Source and contributor context

Dina Genkina · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.

Source media

Media declared by the original publisher; availability checked separately from the editorial reading.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Graphics processing unit

A chip that performs many calculations at once; the main hardware used to train and run AI models.

Benchmark

A defined set of tests used to compare systems. Results depend on the tasks and measures selected.

IEEE Spectrum

IEEE Spectrum reports on engineering practice, research, and the industries around it.

Editorial provenance

Version 1. Publication ID: ad75130e-a16b-46c7-8083-93f8fd7a2d97.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-18T11:34:50.212Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: useful_perspective.

Observation: 42691d0d52e963d3c41725462db22d7855c5d9c02ae8d5c6cebc1e77eb77cc7d. Source text hash: 7b0a6a0882b7490fed02a3559dcd837a50f4e5e95c8b4683d664d19d493e0e9e.

Revision and source-check history

Version 1 · 2026-11-18T11:34:50.212Z

IEEE Spectrum: AI Is Giving Optical Computing Another Shot

’s Dina Genkina describes how pressure to reduce AI inference’s energy use is renewing interest in optical computing: using light to perform calculations. The appeal is that matrix-vector multiplication, a calculation repeated in AI inference, can in principle use less energy with light because the multiplication itself can be passive. But encoding data into light and converting results back to digital form also consume energy, and inference includes operations that optical systems may not handle as well. The article follows several approaches: free-space optics, integrated photonics, and adding calculations to data-center communication paths. Their targets differ, from inference prefill—the phase that processes an input—to decode, which is more constrained by moving data between memory and processors. Company performance figures remain claims or, in one case, results from emulation; they do not establish energy savings in deployed systems. Genkina’s account also cautions against assuming optical computers will replace : the technology may instead complement them or support different computations. The central tension is whether optical systems can deliver useful efficiency at scale despite conversion costs, manufacturing constraints and the portions of workloads they cannot accelerate.

Original source

· Article

Source details

Original source date: Nov 17, 2026 (2026-11-17T14:00:03+00:00).

Date basis: Date stated by the publisher.

Added to Escape Meridian: 2026-11-18T11:34:50.212Z. First observed: 2026-11-17T14:06:48.267Z. Retrieved: 2026-11-17T19:16:06.011Z.

Limits

The article reports company targets and claims, not demonstrated system-wide energy savings at scale. It says the energy advantage of integrated photonics is “yet to be conclusively demonstrated.” LightSolver’s reported speed results are from emulation, and the team is still working on explicitly energy improvements. The article describes potential uses across several optical-computing approaches; it does not establish which approach, if any, will prevail.

Open question

Can optical-computing companies deliver their targeted 10- to 80-fold energy-efficiency improvement at scale before traditional electronics catches up?

Source-stated

Dina Genkina reports that AI inference’s energy demands have renewed interest in optical computing as a possible way to reduce power use.

AI inference—the process of using AI models to respond to user queries—has driven much of the construction of huge, power-hungry data centers

Source-stated

The article explains the proposed energy advantage: optical multiplication can be passive, but encoding inputs and reading results also cost energy, so system size and conversion costs matter.

The only cost is encoding the vector into light and reading out the answer

Source-stated

Optical systems target different parts of AI workloads: some focus on prefill, while Optalysys aims to add computation to communication paths and memory-bound decode.

Lumai, and other companies in the space, are focusing on the compute-intensive prefill phase of inference.

Source-stated

The article reports that optical-computing companies are aiming for a 10- to 80-fold improvement in energy efficiency over today’s GPUs, while asking whether they can achieve it at scale before electronics catches up.

optical-computing companies today are aiming for a 10- to 80-fold improvement in energy efficiency over today’s GPUs.

Source-stated

Genkina presents optical computing as a possible complement to GPUs rather than an assured replacement, with different hardware potentially suited to different tasks.

We see the data center of the future as heterogeneous: CPUs, GPUs and photonic NPUs [native processing units] each take on the tasks they are best suited for

Source and contributor context

Dina Genkina · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.

Source media

Media declared by the original publisher; availability checked separately from the editorial reading.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Graphics processing unit

A chip that performs many calculations at once; the main hardware used to train and run AI models.

Benchmark

A defined set of tests used to compare systems. Results depend on the tasks and measures selected.

IEEE Spectrum

IEEE Spectrum reports on engineering practice, research, and the industries around it.

Editorial provenance

Version 1. Publication ID: ad75130e-a16b-46c7-8083-93f8fd7a2d97.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-18T11:34:50.212Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: useful_perspective.

Observation: 42691d0d52e963d3c41725462db22d7855c5d9c02ae8d5c6cebc1e77eb77cc7d. Source text hash: 7b0a6a0882b7490fed02a3559dcd837a50f4e5e95c8b4683d664d19d493e0e9e.

passed · 2026-11-18T11:34:50.212Z · anthropic/claude-opus-5.5

Publication: ad75130e-a16b-46c7-8083-93f8fd7a2d97

attribution — pass: The reading correctly attributes the feature to Dina Genkina at IEEE Spectrum. It keeps company figures as claims or targets and separates them from the article's own framing. The heterogeneous-data-center excerpt is a quote from Q.ANT's Förtsch, but the claim attributes only the overall complement-not-replacement framing to Genkina, which the article supports.

the computing ecosystem seems ripe for a variety of approaches

summary — pass: The reading leads with the development: AI inference energy pressure is renewing interest in optical computing. It explains computing with light, why passive multiplication might save energy, prefill and decode, and the main caveats (conversion costs, unaccelerated operations, manufacturing). Free-space optics and integrated photonics are named rather than defined, but they are not essential to the central point. The question reflects the article's own stated open question.

The question is whether they can deliver on that promise at scale before traditional electronics catches up to those gains.

date — pass: originalPublished is 2026-11-17T14:00:03Z, taken from the publisher page and matching the feed entry. That is consistent with the page showing '5h' at retrieval around 19:16Z the same day. No earlier publication is indicated.

support — pass: Each claim is source-stated with an apt excerpt. The source supports the limitations: LightSolver's results come from emulation with energy benchmarking pending, and the energy advantage is not yet conclusively demonstrated. Decode being memory-bound and the nonlinear-operation caveat are also supported. No unsupported benefits or costs are implied.

The energy advantage is yet to be conclusively demonstrated, however, and Q.ANT’s Förtsch admits there is still a ways to go.

media — pass: The selected media is the verified direct link to the original IEEE Spectrum article. It is the most direct way for readers to access the source.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.