The Swell, the Break, the Rip, and the Shore — this week in excitonic-photonics, quantum, and the market underneath them.
🌊 Swell — What's Building
The photonic quantum computing market is still in its early build phase but the capital is arriving ahead of the technology: one industry forecast puts the sector at roughly $176 million in 2025, growing to about $5.8 billion by 2035 at a projected 40% compound annual rate, with growth tied to chip design advances, quantum networking, and tighter integration with existing semiconductor fabrication.
That capacity question is showing up in real proposals. Photonic Inc. this week unveiled Project VANGUARD, a proposed CA$500 million (roughly US$359 million) multi-tenant semiconductor facility anchored in Vancouver, aimed at easing fabrication and packaging bottlenecks across quantum computing, AI, and aerospace — part of a broader push toward domestic, shared foundry infrastructure rather than reliance on a handful of foreign fabs.
Elsewhere in the optics supply chain, the funding is smaller but the pattern is the same: NTT and Mitsubishi Electric backed an optical quantum computing startup, and Ayar Labs added another $150 million toward its co-packaged optics build-out. A few weeks earlier, optical connectivity provider iPronics secured new funding to scale operations, and the Canadian government allocated funding for photonic quantum computer developer Xanadu's Toronto fabrication facility. None of this is a single breakthrough — it's the swell before it: money moving toward the unglamorous middle of the stack (packaging, interconnect, fab capacity) that photonic and quantum hardware actually depends on.
The underlying science is quietly catching up to that capital. Physicists at the University of Vienna, led by Andrii Chumak, reported in Science Advances that they'd extended magnon lifetimes to roughly 18 microseconds — nearly a hundredfold jump over the few-hundred-nanosecond ceiling that has limited magnon-based devices until now. Magnons are the collective spin-wave excitations in a magnetic solid, and their appeal for quantum hardware is that they naturally couple to photons, phonons, and other quasiparticles, and can, in principle, be routed on chips as small as those in a smartphone. The team got there by cooling ultra-pure, single-crystal spheres of yttrium iron garnet to 30 millikelvin — cold enough to freeze out the thermal processes that normally kill magnons — and identifying a previously unobserved class of short-wavelength dipole-exchange magnons. The more consequential result was diagnostic rather than record-setting: testing spheres of varying purity, the team found that the remaining limit on magnon lifetime traces to trace material impurities, not any fundamental physical law. That reframes the path forward as a materials-engineering problem rather than a new-physics one — the kind of finding that tends to compound over the next few years of follow-on work, in the same way YIG purity improvements once did for magnonics generally.
🌊 Break — This Week's Main Event
The most consequential technical news of the week came out of the Quandela–NVIDIA partnership. The two companies published a technical white paper outlining an architecture to integrate photonic quantum processing units with classical AI and HPC environments via NVIDIA's NVQLink. The framework links Quandela's Quantum System Controller to NVIDIA GPU nodes over a low-latency interconnect, enabling GPU-accelerated simulation, quantum error correction, and QPU pulse calibration inside a single host process running CUDA-Q and the MerLin quantum machine learning framework.
The significance isn't the demo — it's the plumbing. Hybrid quantum-classical computing has mostly lived in research papers; this is an attempt to standardize how a photonic QPU actually talks to a GPU cluster in production, across everything from cloud experimentation to eventual fault-tolerant deployment. If NVQLink becomes a common interconnect layer the way PCIe or NVLink did for classical accelerators, it changes who can build a hybrid system without reinventing the control stack from scratch — which is the kind of unglamorous standardization that tends to matter more than any single qubit-count headline.
🌊 Rip — The Financial Undertow
The market backdrop this week was more volatile than the technology news suggests. Semiconductor and AI infrastructure stocks sold off sharply mid-week — the VanEck Semiconductor ETF (SMH) fell more than 4% to open options-expiration week — after a coordinated round of public caution from tech leaders, including Anthropic CEO Dario Amodei and Elon Musk, on the pace of AI development, while OpenAI's Sam Altman separately ruled out a 2026 IPO.
The sector recovered by Thursday. The iShares Semiconductor ETF (SOXX) rose about 3.4%, with Intel up 7.7%, AMD up 6.5%, Arm up 8.6%, Nvidia up 2.5%, Marvell up 4.8%, Micron up 5.5%, and SanDisk up 6.2%. The rebound wasn't demand-driven so much as supply-driven: Intel's CEO flagged severe CPU supply constraints and warned that the memory shortage could worsen next year, while reports of exploratory talks between Intel and SK Hynix added to optimism around U.S. chip manufacturing and memory supply — with easing Treasury yields providing a general tailwind across tech.
Underneath the volatility, the earnings picture stayed strong. Micron posted fiscal Q3 revenue of $41.46 billion, up 345.7% year over year and 17.6% above consensus, while Broadcom reported fiscal Q3 revenue of $29.59 billion, up 85.5% year over year. Marvell's expanded custom-silicon program with Google is structured with real skin in the game: the deal carries a warrant letting Google acquire up to 7% of Marvell shares tied to revenue milestones, a detail worth watching for whether other hyperscaler-supplier deals start adopting similar structures.
The net read: this is no longer purely a demand story. Memory and packaging capacity — not chip design or model appetite — is the binding constraint being priced right now, and it's why the "AI slowdown" commentary and the supply-constraint commentary produced opposite stock reactions in the same week.
🌊 Shore — Where It Lands
A few threads worth tracking into next week:
IEEE Quantum Week (QCE26) wrapped in Toronto (Sept 13–18) with roughly 2,222 registrations, 372 technical papers, and 195 posters — a useful proxy for how fast the research community is scaling, independent of any single vendor's roadmap.
Quantum Computing Inc. will showcase an integrated quantum-secure communications platform combining encryption and authentication at ECOC 2026 in Málaga, Spain, September 21–23 — a step toward practical, deployable quantum-secure links rather than lab demonstrations.
Marvell's Investor Day lands October 6 and is the next real catalyst for the custom-silicon and networking growth story that's been driving its multiple.
The bigger picture: with fabrication and packaging now the acknowledged chokepoint in the AI buildout, the photonics-adjacent infrastructure plays — optical interconnect, co-packaged optics, shared foundry capacity — sit closer to the actual constraint than the headline AI names do. That's worth more attention than this week's stock swings, which said more about sentiment than about the underlying buildout.
Disclosure: This newsletter is for informational and educational purposes only and does not constitute financial, investment, legal, or tax advice. Nothing in this issue should be construed as a recommendation to buy, sell, or hold any security, or as an offer or solicitation of an offer to buy or sell any financial instrument. The views expressed are those of the author and are subject to change without notice. Wave.excitonic.com and its authors may hold positions in companies or securities discussed and undertake no obligation to disclose specific holdings or to update this content as circumstances change. Past performance is not indicative of future results, and all investments carry risk, including the potential loss of principal. Readers should conduct their own due diligence and consult a licensed financial advisor before making any investment decisions. Information is believed to be accurate as of the publication date but is not guaranteed and may be sourced from third parties without independent verification.