Murata dates its AI optics bet: first samples this quarter, mass production in 2028, earnings October 30
First samples of Murata's thin-film lithium niobate modulator are due this quarter, with mass production dated 2028. A startup already ships faster parts, so the bet is the ceramic package around them, priced against first-half results on October 30.
Richard Tang · 3 min read
Murata, the Kyoto maker whose business is ceramic passive components, capacitors above all 1, has dated its push into co-packaged optics: first sample of its thin-film lithium niobate (TFLN) modulator this quarter, mass production in 2028, roadmap to 2032 2.
On raw specs it arrives second, so the wager is the package around the part, a second AI revenue leg beyond the passive components it already ships into optical modules 2. That wager now has a price and a date: Murata reports first-half results on October 30 3, with the shares at about 46 times earnings 3.
On bandwidth, Murata arrives second
HyperLight, a TFLN startup, already sells production-grade modulators above the bandwidth Murata is targeting, built for 448-gigabit-per-lane test equipment; the physics is settled and purchasable today 4. Murata aims above 400 gigabits per channel, with 800 promised for 2030 2. The thin-film form of the material is what lets the modulator sit beside silicon photonics at low power 5.
Murata's bandwidth target is the floor of what HyperLight already ships
- Estimate
Data
| Value | |
|---|---|
| Murata target, 2028 (estimate) | 100+ GHz |
| HyperLight 110GHz, shipping | 110 |
| HyperLight 145GHz, shipping | 145 |
The bottleneck is the box around the die
Co-packaged optics (CPO) puts the optical engine beside the switch chip to shorten the copper, cutting signal loss and power 2. What it complicates is everything around the die: power delivery, heat and repair, which get harder once optics and silicon share one package, as Murata itself concedes 2. Its roadmap attacks exactly those problems, from real heritage: the modulator process grows out of its phone radio-filter business 2.
One piece already exists as a physical sample: a 125-millimeter low-temperature co-fired ceramic (LTCC) substrate carrying 400 gigabits per channel 2. Embedded power delivery and cooling join it in 2028, 200-millimeter panels with optical waveguides in 2030, and the last piece, wiring that carries both electrons and light, is dated 2032 and remains a concept 2. Whoever assembles a fast die into a package that survives power, heat and repair collects the value, and few component makers have Murata's ceramic depth to attempt it 1.
Demand is arriving, and the pluggable camp is pushing back
Demand is no longer hypothetical. NVIDIA's+1.57% — NVIDIA, up 1.57 percent today Spectrum-X co-packaged switches entered mass production in August, and TSMC+0.45% — TSMC, up 0.45 percent today, the foundry making the switch chips, has expanded its testing and inspection capacity for the ramp 6. A workstream at the Open Compute Project, the open-hardware standards group where hyperscalers settle rack designs, is standardizing CPO racks, led by Lightmatter$4.63B — Lightmatter, private, latest valuation $4.63B with Celestica+0.27% — Celestica, up 0.27 percent today, Dell, Keysight+1.39% — Keysight, up 1.39 percent today and Qualcomm+0.45% — Qualcomm, up 0.45 percent today signed on 7.
The pluggable camp is pushing back. A Nvidia-Cisco+0.35% — Cisco, up 0.35 percent today specification unveiled this month claims four times the bandwidth density of today's best pluggable modules while keeping them field-replaceable, the exact weakness in Murata's plan, though it also has no ship date yet 8. A ten-year market study lists CPO programs at Coherent and Lumentum, the optical-module incumbents 9.
Two dates turn the roadmap into evidence
Results on October 30 give the first read on what AI demand is earning the capacitor business, in a year when demand for these components has outrun supply 3. The modulator sample is due by the end of December, the first proof the second leg exists outside a lab 2. Beyond it, mass production in 2028 and every hop to 2032, is projection; the far end is explicitly a concept 2.
If the sample lands on schedule, Murata has a credible claim on the package layer of AI networking. If it slips, the multiple stands on capacitors alone.
Get Chips by email, free
How AI demand is reshaping chipmakers, memory and the fabs behind them. In our daily email, every weekday.
