The Modulation Problem: Why We Invested in NLM Photonics

Pangaea Ventures is pleased to announce that we have joined NLM Photonics' Series A2, alongside new investor Diamond Edge Ventures, the venture investment arm of Mitsubishi Chemical Corporation and existing backers Emerald Technology Ventures, Idemitsu, Tokyo Ohka Kogyo, Oregon Venture Fund, and StoryHouse Ventures. David Weekes joins NLM's board. NLM's full announcement is here. Below, why we invested.

Every conversation about AI infrastructure eventually arrives at power, as it should. Most of them stop at the GPU; however, the harder constraint lives another layer down, in the optics that move data between chips, servers, and racks. Some estimates now put networking at up to 50% of all data center power consumption. As models grow and more accelerators are made to work in coordination, the energy cost of moving data is rising faster than the cost of the computation it serves. 

Inside every pluggable transceiver, one component is critical to how far that trend runs: the electro-optic modulator, which encodes data in the form of electrical signals onto light. Silicon photonics got the industry this far on the plasma dispersion effect, which becomes markedly less efficient past 100 Gb/s. Reaching 200 and then 400 Gb/s on a shrinking power budget fundamentally requires new materials.

NLM Photonics, a 2018 spin-out from the University of Washington, has developed a material that precisely targets this bottleneck.

Here’s the full breakdown of why we invested.

Glass, not polymer

Organic electro-optic (OEO) materials are not a new idea. The underlying physics has been understood for decades, and the performance case has always been strong. What sank earlier attempts was thermal stability: polymer platforms lose chromophore alignment under heat, and the field of OEOs acquired a reputation that stuck, fairly or not.

Selerion™, NLM's proprietary material, is a crosslinked thermoset glass rather than a polymer. Small-molecule precursors are deposited into a slot waveguide and then poled under an electric field while heated, which aligns the chromophores and crosslinks them into a dense network in the same step. The alignment is locked in place chemically.

The results are what convinced us. Selerion™ devices have run several thousand hours at 85°C with no measurable degradation, have passed damp heat testing at 85°C and 85% relative humidity, part of the Telcordia GR-468 qualification sequence, and project beyond a decade of operating life. Modulation efficiency at 200 Gb/s is more than ten times that of standard silicon photonics, verified by third parties on production wafers. No prior organic electro-optic material has reached this standard at the device level.

Beyond these results, the footprint and energy use are highly appealing. Because of the modulation physics implicit to NLM's materials, the interaction length is a few hundred microns against 5 to 10 mm in silicon. The die is 40% smaller at 20 to 50% lower energy per bit.

The material goes on last

The second reason is about when the material is integrated into a device. Thin-film lithium niobate (TFLN), indium phosphide, and barium titanate all offer good electro-optic physics; but each requires a foundry to build dedicated front end capacity. NLM's material is applied after a standard silicon photonic wafer run is complete, using a surface activation, deposition, encapsulation, and poling process that invokes no front-end modifications.

This means fabs don’t need to worry about contamination of other materials, and because of that, are willing to adopt the solution that NLM has built: Their 1.6 Tb/s transmitter PIC, eight lanes at 200 Gb/s, was designed with Enosemi (acquired by AMD), fabricated at AMF (acquired by GlobalFoundries) on standard 200mm wafers, delivered to Centera Photonics for a Tier 1 AI customer, and demonstrated at OFC 2026. A 400 Gb/s per lane version is taped out and currently in sampling with other customers.

Fabs don't buy from competitors

The optical supply chain runs on trust built over decades between foundries, OEMs, design houses, and materials suppliers. A startup that arrives selling chips is asking those companies to take on enormous risk. NLM’s business model shrewdly avoids this scenario by licensing materials, process, and device IP instead. This approach is far more compatible with the roadmap of a risk-averse customer or partner.

Customer to CEO

Brad Booth spent most of his career on the buying side of this transaction. At Microsoft he was principal network architect in the Azure hardware architecture group. At Meta he worked on optical connectivity for cloud and AI data centers, and he has spent years in the standards bodies where transceiver roadmaps get argued out.

He knows what hyperscalers will specify, roughly when they will specify it, and which claims they will not accept without independent data. 

Technical continuity runs through co-founder and CTO Lewis Johnson, who did the underlying chemistry for his PhD at the University of Washington and co-invented the foundational patents. Johnson works from chromophore synthesis through to system-level PIC design, and the deposition scale-up has advanced on his direct involvement.

What the buyers are paying for

Silicon's limits are felt most acutely by the companies furthest downstream, to the extent that we have seen many of them act on it. Marvell bought Celestial AI for $3.25B while projecting meaningful revenue only in fiscal 2028. AMD bought Enosemi, Molex bought Teramount, Marvell also took Polariton. These are bets based on the potential for long term value creation in a hot and fast moving market, priced years ahead of the revenue. Our belief is that NLM will become an equally crucial acquisition target for the build-out of photonics once they prove out the reliability of their material.

Where this sits in the stack

Photonics and Networking is one of six sectors in The Hard Tech Report 2026, our joint research with Hakuhodo DY Holdings, and it contains one of the highest concentrations of segments we score as Transformative. Co-packaged optics (CPO) is among them, with TSMC's COUPE platform entering mass production this year. CPO puts a hard premium on compact, power-efficient modulation, which is where a few hundred microns of organic glass has a structural advantage over millimeters of inorganic crystal.

The report also argues that Japanese materials suppliers are well placed to lead at this layer. NLM's investor base reflects that: TOK, a leading photoresist manufacturer, has backed the company since 2023 and is collaborating on market-ready inks carrying NLM's materials, alongside Hamamatsu Photonics and Idemitsu.

We score opportunities on Transformation Potential: the magnitude of the performance leap, the value density of the layer it lands in, and how irreversible the position becomes once adopted. The modulator sets what each lane can carry, so value concentrates there. And a process module written into a foundry PDK is not casually removed; every design built on it would have to be redrawn.

Irreversibility is the one that argues for moving now. The industry is deciding what carries it past 800 Gb/s per lane, and those decisions get locked into process kits and product roadmaps for years. There is also a straightforward arithmetic to the energy case: 20 to 50% less power per bit, applied across hundreds of millions of transceivers a year, is a real reduction in what AI infrastructure draws from the grid.

NLM has the reliability data, a customer PIC, a business model the incumbents can say yes to, and a team that knows both the chemistry and the buyer. We are glad to be backing Brad, Lewis, and the rest of the team.