On October 29, 2026, CollPlant Biotechnologies (Nasdaq: CLGN) announced that its subsidiary LightSolver has signed an agreement with the Directorate of Defense Research & Development (DDR&D) within the Israel Ministry of Defense for the development and delivery of its Laser Processing Unit (LPU), a pure photonic computer. The agreement follows earlier successful proof-of-concept work with DDR&D and establishes a multi-year framework engagement, with an initial order covering the first 12 months. According to the release, it was formally signed by senior representatives of the ministry and includes provisions governing intellectual property and technology usage rights.
What the announcement does not contain is almost as informative as what it does. There is no contract value, no performance figure for the LPU on any workload, no named defense application, and no delivery date for a system. Readers should hold both halves of that picture at once: a government defense customer has moved from a proof of concept to a paid, multi-year framework, which is a real commercial milestone; and the public evidence for what the machine can do has not advanced by a single number.
That distinction — customer validation versus capability demonstration — is the right lens here. A customer signing a development agreement is evidence about a customer’s judgment and priorities, not, by itself, evidence about a machine’s speed, accuracy, or energy efficiency.
What was actually announced
The parties and the structure are straightforward, and all of them are verifiable from the announcement itself. LightSolver, founded in 2020 by Weizmann Institute physicists Dr. Ruti Ben Shlomi and Dr. Chene Tradonsky, builds what it calls the Laser Processing Unit: a computer whose core is a closed-loop, all-optical system that maps computational problems onto the physical dynamics of interacting lasers, keeping the iterative compute loop in the optical domain rather than shuttling intermediate results back into electronics at each step. CollPlant, a company previously known for plant-derived recombinant human collagen for regenerative medicine and medical aesthetics, completed its acquisition of LightSolver in September 2026.
The DDR&D — known in Israel by its Hebrew acronym MAFAT — is the Defense Ministry’s research and development directorate. Its Head of Emerging Technologies Branch, quoted anonymously in the release, said LightSolver’s architecture “offers a fundamentally different approach to addressing demanding computational challenges.” Ben Shlomi said the company was “proud of being chosen by the Israel Ministry of Defense to assist in solving advanced-computing problems it is evaluating,” and pointed to the engagement alongside LightSolver’s existing multi-year engineering program with Boeing as evidence of converting “successful technical validation into paid customer programs.”
The framework structure matters. A multi-year framework with a funded first 12 months is not a one-off purchase, and it is not a production deployment either. It is a development and evaluation engagement, with future phases presumably contingent on results — a contingency the release’s own forward-looking statements section makes explicit, warning that DDR&D may not fund future phases, or that the program could be delayed, reduced, or terminated. The honest reading is that LightSolver has earned a funded seat at the table for the next year, not that it has won a finished product contract.
One clarification is worth making early, because secondary coverage has already blurred it. The release’s headline figure — a defense photonic-computing market “estimated at up to $5 billion annually” — describes an estimated total addressable market, attributed in a footnote to Hyperion Research 2025, not the value of this agreement. No contract value is disclosed anywhere in the announcement, and the figure itself should be treated as vendor framing: a company-cited estimate, not an independently verified measurement.
Why a defense customer is a different kind of signal
Photonic computing companies have spent the last few years accumulating laboratory demonstrations, often impressive, of optical systems performing some computation faster or at lower energy than a digital baseline on a narrowly defined task. What they have accumulated much less of is evidence that a demanding customer will pay, repeatedly, to develop the technology for its own problems. That is why the DDR&D agreement matters more than its thin technical content might suggest.
Defense customers are unusually demanding evaluators in one specific respect: their workloads are not benchmarks chosen to flatter the hardware. A defense R&D directorate evaluating “advanced-computing problems” is presumably bringing problems it actually needs solved, with requirements set by its own engineers rather than by the vendor’s marketing department. Surviving a proof of concept in that environment and converting it into a paid framework is a stronger signal than surviving a laboratory demonstration, because the customer controls the test.
It is also a signal about the kind of computing LightSolver is selling. Most photonic AI coverage on this desk concerns neural network computation: the linear algebra of AI inference and training. LightSolver’s LPU is aimed at a different class of problem — high-dimensional mathematical workloads: optimization problems and the large, tightly coupled systems of partial differential equations that underpin physics-based simulation, where the computation is the physics itself, not a neural network approximating it.
This distinction explains the Boeing program, the other paid engagement Ben Shlomi cited. Announced in April 2026 as a strategic financial partnership, the Boeing collaboration funds continued LPU development for complex engineering simulations, with an emphasis on degradation-driven structural effects — the slow physical processes that determine how long a material or component lasts. Boeing’s interest, as stated in that announcement, is in numerical accuracy, repeatability, and integration with existing high-performance computing environments: the unglamorous requirements that separate a laboratory curiosity from a tool an engineering team will actually use.
The DDR&D engagement fits the same pattern. Whether the LPU can serve a defense ministry’s simulation and optimization problems is still an open question — but the fact that two large, technically sophisticated customers in aerospace and defense have now paid to find out is itself a data point about where photonic computing’s near-term market may actually be. It may be in solving equations, not in running chatbots.
How the LPU differs from the photonic AI field
The dominant commercial approach to photonic AI builds chips that perform matrix multiplication optically, while nonlinear activation functions and control remain electronic. Those systems are, in the honest framing this desk has maintained, hybrid accelerators: optics for the linear part, electronics for everything else, with system-level energy and accuracy determined by the whole pipeline rather than the optical core alone.
The LPU claims something more radical, and readers should weigh the claim accordingly. Instead of using optics to accelerate one operation inside a digitally controlled pipeline, LightSolver says the iterative computation itself stays in the optical domain: a problem is encoded onto interacting lasers, and the lasers’ own dynamics perform the search for a solution. The “closed loop” in the company’s description is the point: intermediate results are not converted back to electronic signals, processed, and re-encoded at each iteration, which is where hybrid systems typically lose much of their optical advantage to conversion overhead. Until LightSolver publishes a full-pipeline measurement, “all-optical core” should be read as a description of the compute loop, not of the whole machine.
That premise carries a characteristic trade-off. A system whose computation is its physics is constrained by what that physics can express, by how precisely the problem can be encoded onto it, and by how accurately the answer can be read out. The questions that determine whether the LPU is useful are therefore mundane-sounding: problem size, solution quality and repeatability, and integration into an existing high-performance computing workflow. The Boeing announcement’s emphasis on numerical accuracy and repeatability suggests LightSolver’s customers are asking exactly these questions. The DDR&D announcement answers none of them publicly.
The scrutiny file
A careful reader of this announcement should carry a short list of what remains unproven, because the release’s confident tone can easily outrun its contents.
First, no performance evidence is offered. There is no benchmark, no problem size, no time-to-solution, and no energy figure in the DDR&D announcement. Earlier company materials cited in trade coverage have mentioned projected accelerations for specific simulation workloads — figures derived from emulation and proof-of-concept studies, not from a delivered system running a customer’s problem — and those projections should be treated as modeling results awaiting hardware confirmation, not as findings.
Second, no application is named. “Advanced, compute-intensive applications” and “complex computing workloads” are placeholders, not descriptions. Defense work is legitimately secretive, and the absence of a named application may reflect classification rather than vagueness — but the reader cannot distinguish those two explanations from the outside.
Third, the commercial timeline is long and the milestones are company-set. Trade coverage of LightSolver’s roadmap has described proof-of-concept and minimum-viable-product systems targeted for 2027 and a standard system for 2028, with the usual caveat that timelines in novel hardware slip more often than they hold.
Fourth, the corporate context is unusual and deserves a plain statement. LightSolver is now a subsidiary of a Nasdaq-listed regenerative medicine company. That is not a criticism — the acquisition gives LightSolver access to public-market capital — but readers evaluating its claims should look for technical publications and customer evidence, not only for parent-company press releases, because the press-release channel is structurally optimistic.
None of this amounts to skepticism about the physics. The point is narrower: the DDR&D announcement proves that a sophisticated customer will pay to develop the technology. It does not prove the technology works at the scale, accuracy, or efficiency its customers need. Those are different claims, and conflating them is how photonic computing hype cycles get started.
Where this fits, and what to watch
Set against the broader photonic computing landscape, the DDR&D agreement is part of a visible shift in who is buying. The field’s early commercial story was about AI inference: startups building optical accelerators for datacenter AI workloads, competing on energy per operation against GPUs. That story has not disappeared, but it has been joined by a second one, in which the buyers are engineering and defense organizations with simulation and optimization problems that GPUs handle poorly or expensively.
Israel’s own ecosystem adds context. A national quantum ecosystem report published by the Israel Innovation Authority and a World Economic Forum-affiliated center in late October 2026 describes LightSolver’s optical computing as “quantum-inspired,” while noting that the Directorate of Defense R&D was one of the funding bodies behind the first phase of Israel’s national quantum initiative. The DDR&D agreement is consistent with that picture: a defense R&D establishment treating an optical computer as a strategic capability worth developing domestically, rather than waiting to buy it from abroad.
For the superintelligence angle this desk tracks, the implication is indirect but real. Advanced AI depends not only on training and inference but on simulation, optimization, and scientific computing — the workloads LightSolver targets — and on sovereign capacity to build computing hardware outside the GPU supply chain. A defense ministry investing in an all-optical solver is betting that the future of consequential computing is broader than neural networks. That bet may prove right or wrong. The fact that it is now a funded program, rather than a laboratory proposal, is what is new.
Five developments would turn this announcement from a validation signal into a capability story: a named problem class with stated results on LPU hardware, against a digital baseline; a peer-reviewed publication or technical report on the LPU’s architecture and behavior on a real problem; evidence of repeatability across runs and, eventually, across more than one system; movement on the framework itself, such as a funded second phase or follow-on order; and independent coverage — a benchmark run by a customer or third party, rather than syndication of the company’s own release.
Until those appear, the fair summary is this: Israel’s defense R&D directorate has paid to develop an all-optical computer, after a successful proof of concept, inside a multi-year framework whose first year is funded. That is a meaningful commercial milestone, a meaningful signal about where photonic computing’s early customers are, and no evidence at all — yet — about how well the machine computes. A paid defense program is a step toward closing that gap, and it is worth following for exactly that reason.
Sources and further reading
The core facts — the DDR&D agreement, its multi-year framework structure, the funded first 12 months, the prior proof-of-concept work, the intellectual-property and usage provisions, and the quoted remarks — come from the announcement issued by CollPlant Biotechnologies via GlobeNewswire on October 29, 2026. The $5 billion annual figure is the release’s own estimated defense-sector total addressable market, footnoted to Hyperion Research 2025; no contract value is disclosed, and the figure is not the value of this agreement. Background on LightSolver’s founding in 2020 by Ben Shlomi and Chene Tradonsky and its Boeing partnership comes from the April 2026 Boeing–LightSolver announcement and contemporaneous Photonics Spectra and HPCwire coverage. CollPlant’s completion of its LightSolver acquisition in September 2026 is as reported in company announcements and trade coverage. Context on Israel’s quantum and photonic ecosystem comes from the Israel Innovation Authority report published in late October 2026. For background on the hybrid reality of photonic systems and the distinction between device results and proven systems, see this desk’s earlier coverage of photonic AI accelerators and the MIT–Neurophos collaboration. Readers seeking performance benchmarks for the LPU should look for future peer-reviewed publications or customer-run evaluations; none were available at the time of writing.