--- title: "NVIDIA, which loves copper the most, why did it suddenly start working on light?" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/41853715.md" description: "At that launch event, what Jensen Huang pulled out wasn't a GPU. It was two switches—and two optical switches at that: Quantum-X and Spectrum-X Photonics, directly soldering light into the switch chip (industry jargon calls it co-packaged optics, CPO). The time was the 2025 GTC. At that moment, only one sentence came to my mind: the people selling shovels have started digging for gold themselves. Over a year has passed, and this hasn't become old news—it's just starting to accelerate, the real main event is still to come. This has to be connected to the dividing line from the previous article..." datetime: "2026-06-13T12:45:03.000Z" locales: - [en](https://longbridge.com/en/topics/41853715.md) - [zh-CN](https://longbridge.com/zh-CN/topics/41853715.md) - [zh-HK](https://longbridge.com/zh-HK/topics/41853715.md) author: "[MingPanda](https://longbridge.com/en/profiles/16386316.md)" --- # NVIDIA, which loves copper the most, why did it suddenly start working on light? At that press conference, what Jensen Huang pulled out wasn't a GPU. It was two switches—and two optical switches: Quantum-X and Spectrum-X Photonics, welding light directly into the switch chips (industry jargon calls it co-packaged optics, CPO). The time was GTC 2025. At that moment, only one sentence flashed in my mind: The shovel seller has started mining himself. More than a year has passed, and this hasn't become old news—it's just starting to accelerate, the real main event is still to come. This needs to be connected to the dividing line from the previous article. Most people didn't take it seriously at the time because everyone thought NVIDIA was just "a graphics card maker." The key point isn't "NVIDIA added a new product line" The real weight of this matter can only be understood by placing it back within the context of line ③. Last time we said: that NVIDIA which kept stuffing 5,000 copper cables into server racks and publicly stated "copper is the best connection method"—it was copper's number one die-hard user. Now, this number one die-hard user of copper has entered the optical field itself. This is the inflection point. It's not some analyst report predicting "a future shift to optics"; it's the person defining this industry chain writing optics into his own product roadmap. But where it cuts first is very deliberate NVIDIA didn't start by ripping out all the copper inside the racks. It was restrained—the first cut was "between racks." The NVLink backplane with 5,000 copper cables inside the rack hasn't been touched yet. Why? Precisely because of that dividing line ③: Where copper can still reach (that short segment inside the rack), continue using copper—cheap, power-efficient, robust; where copper can no longer reach (between racks, between data centers), it's optics' turn to take the stage. Looking at its future roadmap, the rhythm is the same: the next-next generation (Rubin) will start discussing "opting for optics inside the rack" for some models; the even more distant generation (Feynman, around 2028) will simply give customers the choice of "copper or optics." Copper isn't being killed with a single stroke. It's this dividing line, moving generation by generation deeper into the rack. Why now, of all times Three unavoidable forces pushed NVIDIA across this line together. Can't reach. What it needs to connect is no longer 72 GPUs, but "millions, across data centers." At this scale, copper is simply out of the game outside the rack, only optics can work. Too power-hungry. Filling an entire data center with traditional optical modules consumes a staggering amount of power. Welding light into the switch chip, according to NVIDIA's own claims, brings significant efficiency gains and a drastic reduction in laser count—all the saved power goes back to computing. (These multiples are NVIDIA's official comparison data; take the direction, not as independent conclusions.) Too many things to break. Scaling out to millions of GPUs, the number of optical modules and lasers becomes astronomical, each a potential point of failure. Fewer components, the less likely the entire network is to fail. Can't reach, too power-hungry, too many things to break—three issues hitting their critical points almost simultaneously. So NVIDIA didn't wait, it entered the field itself. Therefore, this article is the signal of "the industry starting to act" ①There are two ways for GPUs to die. ② An AI cluster isn't a pile of GPUs, but a bandwidth network. That's about perception. ③ Everyone says AI needs to switch to optics, but NVIDIA stuffed in 5,000 copper cables. That's about principles. In this article, the story shifts for the first time from "theory" to "action"—a real company, a real point in time (GTC 2025), a real product (Quantum-X, Spectrum-X Photonics). Optical interconnect is no longer "will happen," it's "already happening." That dividing line is starting to shift massively towards the optics side. When the line moves, money follows. Where will it land—on the switch chips, the optical engines, or the lasers? That's the calculation for the next article. Before the reveal, I'd like to hear yours first—which segment are you betting on? Leave a comment, and we'll compare answers in the next one. $Marvell Tech(MRVL.US) $Coherent Corp.(COHR.US) $Lumentum(LITE.US) ### Related Stocks - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [COHR.US](https://longbridge.com/en/quote/COHR.US.md) - [LITE.US](https://longbridge.com/en/quote/LITE.US.md) - [MRVL.US](https://longbridge.com/en/quote/MRVL.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [07788.HK](https://longbridge.com/en/quote/07788.HK.md) - [07388.HK](https://longbridge.com/en/quote/07388.HK.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md) - [NVDQ.US](https://longbridge.com/en/quote/NVDQ.US.md)