Copper wiring still carries almost every bit of data moving between GPUs in a modern AI cluster, and Lumilens just raised $700 million betting that stops being true (Lumilens). The San Jose company is two years old. It has now raised more than $900 million total, at a valuation of $5.51 billion (Lumilens). The new round was co-led by Atreides Management, Bain Capital Ventures, Meritech, Seligman Ventures, and Spark Capital, with more than a dozen other firms joining, including Qualcomm Ventures and J.P. Morgan Private Capital.

Lumilens is already shipping equipment into a hyperscale cloud provider's data centers under an agreement worth billions of dollars over several years. The company declined to name the customer (Gardizy).

A Repeat Founder Bets on Connectivity, Not Chips

Chief executive Ankur Singla has sold two companies before this one. Contrail Systems went to Juniper Networks, Volterra went to F5. His technical co-founder, Ted Schmidt, spent years as a distinguished engineer at Juniper building on the silicon photonics work Juniper acquired through Aurrion back in 2016. That pedigree is why investors wrote checks into a two-year-old company at a $5.5 billion price. It is not why the company will win.

Lumilens builds two product lines. One connects GPUs within a single rack, the scale-up network. The other stitches racks and clusters together, the scale-out fabric. As clusters grow past hundreds of thousands of GPUs, both networks strain, and copper strains first over distance.

Marvell Already Cashed Out of This Bet

Four companies, four different outcomes, same underlying wager. Marvell acquired Celestial AI in February for $3.25 billion, before Celestial AI ever had to prove it could raise at that price independently (Gardizy). Lightmatter, founded in 2017, has raised $850 million and was last valued at $4.4 billion in 2024 (Gardizy). Ayar Labs took strategic capital directly from Nvidia and AMD rather than chase a standalone valuation at all (Bellamkonda, "The End of Copper").

Lumilens took the fourth path: pure financial capital, no strategic chip-maker anchor, and a valuation set by growth investors who need an exit. That is a different bet than Ayar Labs took, and a riskier one than Celestial AI's.

Singla Admits Copper Still Wins on Price

"Copper is way cheaper today and will be way cheaper than optics." (Gardizy)

Singla said that in an interview, and it is a rare thing for a founder to say out loud about his own category. Optical gear can cost several times what copper costs for the same connection, and Singla's stated goal is closing that gap through manufacturing scale, not through better chip design alone (Gardizy). That has pushed Lumilens to spend as much effort on how it builds its products as on what those products do.

Seligman Ventures partner Umesh Padval, an investor in Lumilens, put it more bluntly: optics has existed as an option for years without displacing copper, and the AI boom may finally force the switch he has been waiting for (Gardizy).

A Pending China Import Ban Turns Manufacturing Into Strategy

Singla has also committed to manufacturing outside China. That commitment picked up weight this week. Reuters reported the Trump administration is weighing a ban on imports of certain data-center components from China, optical gear included (Gardizy). If that ban lands, the vendor with domestic or allied manufacturing capacity already built stops competing on cost alone. It starts competing on who can ship at all.

Unknowns and Uncertainties

The unnamed hyperscaler customer matters more than the funding round. A multibillion-dollar agreement with one buyer is not the same as broad market validation across the handful of companies that actually build frontier-scale clusters.

Whether $5.51 billion reflects revenue Lumilens has actually booked, or the same growth-investor optimism now pricing every AI infrastructure company at a premium, is not something a Series C press release settles.

The scope of the reported China import ban is not final. Whether it names optical transceivers specifically, or catches them incidentally inside a broader components category, changes how much of an edge Lumilens actually holds.