Nvidia’s efforts to shore up its own supply chain may just increase lead times for other companies, as reports suggest the chip giant has snapped up significant capacity of electro-absorption modulated lasers (EML).

TrendForce reports that Nvidia has purchased EMLs for its own machinations to such an extent that rival optical module makers could face wait times for components beyond 2027, in what could even lead to a worldwide shortage.

An EML is a specialized, high-speed light source used in fiber optic communications that efficiently sends large amounts of data over long distances. They’re made up of two parts: a laser that creates a continuous, pure beam of light and a modulator unit that encodes the data. When brought together, despite their compact size, EMLs can support fast data transmission over long distances – making them vital for use in high-speed optical communication systems, including in data centers to boost network performance and efficiency.

Like other parts of the chip semiconductor supply chain, there’s only a handful of suppliers, given they’re highly complex to produce.

Nvidia has already signed deals with a handful, including Lumentum, which is one of the largest laser manufacturers in the world and is supplying laser technologies for photonic iterations of its Spectrum-X networking switches. Nvidia is also working with Coherent (the parent company of optical component maker Finisar).

That leaves a handful of companies for rivals to pick up their own EMLs, including Macom, Sumitomo, and Broadcom, which offers its own photonic-based networking chips.

Nvidia’s reported effort to shore up its supply chain comes as it’s set to launch its long-awaited co-packaged optic (CPO) offerings in 2026. Revealed at GTC back in March, both its Spectrum-X Ethernet and Quantum-X InfiniBand networking fabrics are set to gain support for silicon photonics switches next year.

According to TrendForce, Nvidia’s silicon photonics and CPO efforts have advanced more slowly than anticipated, with the company moving to pre-allocate a sizable amount of EML capacity to ensure supply.

But while shoring up its own supply, Nvidia’s EML moves may force an industry-wide rush for capacity among non-Nvidia players.

It’s a predicament predicted by McKinsey earlier this year. A forecasting report published in July suggested that the rise of AI-enthused networks could well lead to supply shortfalls.

That report pointed the finger of blame at hyperscalers for any potential optical-transceiver shortage, claiming the biggest players will look to shift 87% of their back-end optics to 800 Gb/s by 2029, with most players currently leveraging 400 Gb/s offerings. And for the eventual rise of 1.6 Tb/s transceivers, McKinsey suggests hyperscalers will account for more than 40% of related demand.

McKinsey’s report advised the networking optics industry to “work quickly and collaboratively” to address a potential laser supply shortfall – though if TrendForce’s reporting is correct, Nvidia may have just jumped the gun on that.

First memory, now lasers?

The reported laser dearth adds to growing concerns about the world’s memory capacity.

Recent weeks have seen memory suppliers like Samsung raise their prices by as much as 60% amid an influx of demand and potential panic buying after OpenAI secured a deal with Samsung and SK Hynix for what equates to 40% of global DRAM output.

In the weeks since, the memory panic has already seen HP admit rising prices could affect its profitability in the second half of fiscal 2026, while Hewlett Packard Enterprise (HPE) had its ratings cut by Morgan Stanley over concerns that OEMs face gross-margin compression due to rising memory costs.

While it’s shored up its laser supplies, reports last week suggested Nvidia was looking to unbundle video RAM (VRAM) from its GPUs for the consumer and gaming market. Such a move would force add-in card (AIC) partners to source their own memory, leaving them to foot the bill amid soaring prices. No official confirmation on this has been made, however.

The memory move could impact Nvidia in other ways, though, with SK Hynix reportedly delaying production of its next-generation memory chips. Those HBM4s chips were due to be part of the upcoming Rubin GPU line, with production bushed back to the third quarter of 2026, instead of the second. Nvidia has, however, denied claims that Rubin could face a production delay, revealing that taping out of the design (the final design process prior to production) occurred back in August.