Download the Complete Report Here
Optical-Loss Progress and Higher Wafer Throughput Advance Hardware Roadmap as Growth Investment Accelerates
Key Takeaways:
XNDU advanced hardware execution, achieving 0.085 dB edge-coupling loss while increasing thin-film lithium niobate and silicon nitride fabrication by 75% and 50%.
Higher wafer throughput accelerates qubit-factory iteration, with the aggregate optical-loss gap reduced by up to 200x and 5-10x remaining.
PennyLane expanded through QROM, Frontier, Lockheed Martin training, and bank engagements, strengthening the developer funnel while monetization remains early.
Project OPTIMISM is nearing a potential update, while $312.8 million of cash and $67.2 million of ATM proceeds support faster roadmap investment.
Valuation remains milestone-driven, with rerating dependent on further loss reduction, qubit-factory progress, funding conversion, and partner monetization.
XNDU’s 2Q26 hardware progress and higher fabrication activity are beginning to translate post-listing capital into faster roadmap execution. XNDU’s second public-company quarter provided more tangible evidence that its expanded capital base is supporting measurable photonic-component improvements and a faster development cadence. The company achieved average edge-coupling loss of 0.085 dB per facet, supported by its internal advanced photonic chip-packaging facility, customized fiber and fiber-array work with Corning, and wafer-singulation support from DISCO. The 0.085 dB result represents a significant component-level improvement and, according to XNDU, may be the lowest edge-coupling loss achieved in the industry.
Lower edge-coupling loss strengthens hardware efficiency, but broader progress toward fault tolerance still depends on reducing losses across the full system. Each transition between fiber and photonic chips introduces loss, making lower coupling loss increasingly important as XNDU connects more chips and modules. Edge coupling, however, is only one of approximately 16–17 major loss contributors across the end-to-end system, alongside qubit generation, beam splitters, phase shifters, ring resonators, optical propagation pathways, gates, and detector efficiency. The 0.085 dB result therefore represents an important component-level milestone, but not a standalone measure of total system readiness.
XNDU’s remaining 5-10x loss gap provides a more measurable framework for tracking progress toward fault tolerance. The aggregate optical-loss gap has declined by as much as 200x over roughly four years, with a further 5-10x reduction still required to reach the threshold for scalable fault-tolerant operation. This gives investors a clearer benchmark for evaluating whether future hardware improvements are translating into meaningful progress against the company’s long-term roadmap.
XNDU is addressing the remaining loss gap through both hardware improvement and architectural simplification. Progress is coming from more chip runs and improvements across fabrication, packaging, propagation loss, coupling, and detector efficiency, while architectural changes are designed to reduce the number of components and optical operations photons must traverse. XNDU expects to provide a more detailed loss and hardware roadmap around the end of summer or Analyst Day, consolidating these contributors into two principal optical paths and extending the framework toward 2029-2030. The update should be an important near-term catalyst by providing clearer benchmarks around current performance, required thresholds, timing, and system dependencies.
Higher wafer throughput should accelerate development cycles across the company’s two core material platforms. Fabrication activity increased approximately 75% for thin-film lithium niobate and 50% for silicon nitride. Silicon-nitride availability through NY CREATES increased from approximately 100 to 150 wafers, while thin-film-lithium-niobate activity through WaferTech, a UMC subsidiary, increased from slightly above 100 to approximately 175 wafers per month. XNDU expects both corridors to increase further.
The higher cadence should improve iteration speed and increase the productivity of R&D investment. More wafer availability allows the company to run additional design, fabrication, testing, and redesign cycles in parallel, helping identify process limitations faster and feed improved components into the qubit-factory architecture. Combined with XNDU’s foundry-compatible approach and access to established semiconductor and photonics infrastructure, this should improve repeatability, process control, and throughput while reducing execution risk at each development cycle.
Albany expansion should strengthen XNDU’s access to U.S. semiconductor infrastructure and engineering talent. XNDU is expanding its U.S. operations around Albany, New York, where proximity to semiconductor research, photonics infrastructure, foundry partners, and government stakeholders should improve coordination across design, fabrication, packaging, and testing. U.S. headcount has increased more than fivefold since 2023, with further significant growth expected by year-end. The public listing has also improved recruiting visibility and compensation flexibility through liquid equity awards, broadening the pool of specialized engineering candidates. Together, the expanded U.S. footprint and deeper talent base should support higher tapeout and wafer activity, greater parallel component development, and a faster hardware iteration cadence.
QROM improvements demonstrate how software innovation can reduce future hardware requirements and reinforce the value of the full-stack model. The company published a patent-filed quantum read-only memory, or QROM, technique that cuts required Toffoli-gate operations by roughly half by reducing unnecessary data-movement and unlocking steps. Because QROM is a common subroutine for loading classical data into quantum algorithms, lower gate requirements could ultimately reduce physical-qubit, error-correction, and runtime needs, easing the burden on the hardware roadmap; the implementation is already available through PennyLane. XNDU also trained a Fourier-based quantum-machine-learning model with more than one million parameters to learn the distribution of ribosomal RNA, demonstrating the software stack’s ability to support increasingly complex scientific workloads, although commercial applicability remains early.
