The Optical Signal: Decoding Goldman Sachs’ AI Bet Through a Crypto Lens

Goldman Sachs just lit a fuse under a Chinese optical module maker called Zhongji Xuchuang. Their new report projects profit growth of 65%, 108%, and 119% for 2026, 2027, and 2028 respectively. That’s not a typo. It’s a bet that AI infrastructure will expand so relentlessly that the humble fiber optic transceiver becomes the new silicon. I’ve spent my career hunting narratives in digital assets—tracing sharding roots, mapping liquidity flows. But when a blue-chip bank makes a call this loud on a hardware supplier, I listen. Not because I trust their models, but because the signal they’re decoding is the same one that drives the entire tech ecosystem, including crypto. Let’s strip the jargon. Optical modules convert electrical data into light pulses and back again. They connect the GPUs in a training cluster. Without them, you can’t scale AI beyond a few cards. The jump from 800G to 1.6T (and then to 3.2T) means each module can carry more data per second. That’s essential for the massive parallel compute jobs behind large language models. Goldman is saying that demand for these modules will follow a hockey stick curve because AI cluster buildouts are accelerating. But this narrative isn’t just about AI. It’s about any system that needs high-bandwidth, low-latency communication between distributed compute nodes. That includes blockchain validators, layer-2 sequencers, and decentralized physical infrastructure networks (DePIN). Every time a crypto minter or a zk-rollup sends a batch of proofs, it relies on similar hardware. The difference is scale and price sensitivity. AI clusters buy 1.6T modules at $10,000 each. Crypto networks buy 100G modules at $500 each. But the underlying trend is the same: the network becomes the bottleneck, and optical interconnect becomes the release valve. I remember 2017, when I was reverse-engineering Zilliqa’s sharding whitepaper. The founders told me that the biggest challenge wasn’t the consensus algorithm—it was the network latency between shards. They were already thinking about optical interconnects. Back then, it seemed futuristic. Today, it’s the default. That experience taught me to look at infrastructure stories with a skeptical eye. Hardware companies often enjoy temporary monopolies during technology upgrades. But the real question is: who captures the value? Goldman’s analysis points to Zhongji Xuchuang as the ‘TSMC of optical communications.’ That’s a powerful metaphor. TSMC doesn’t design chips; it manufactures them at scale with insane yield. Similarly, Zhongji is a manufacturing powerhouse. They turn laser diodes and silicon photonics into modules that work reliably in hyperscale data centers. Their moat is not a breakthrough invention—it’s relentless process engineering and customer trust. They supply Nvidia, Microsoft, Google, and Amazon. That’s a network of relationships that takes years to build. But here’s where the narrative gets tricky. The crypto world has seen this movie before. In 2020, during DeFi Summer, I tracked 50 Uniswap liquidity providers and found that 80% lost money to impermanent loss while chasing APY. The hype was real—but the underlying economics were fragile. Similarly, the optical module narrative seems bulletproof today. Everyone wants to own the picks-and-shovels of AI. But the data tells me to look for the cracks. First, the cycle risk. AI capital expenditure may not grow at 65% per year indefinitely. We’re already seeing diminishing returns from scaling model parameters. If training efficiency plateaus or if a cheaper alternative (like LoRA or quantization) reduces the need for massive clusters, then demand for 1.6T modules could soften. Second, the self-build threat. Microsoft is developing its own optical interconnect technology called Lyra. Google has similar projects. If hyperscalers bring module design in-house, they’ll squeeze suppliers like Zhongji. The same vertical integration that killed many GPU server makers could hit optical module vendors. I’ve been mapping the digital tribe’s hidden rhythm for years. I notice when a narrative becomes too tidy. Goldman’s note is a perfect example. They offer a linear projection of exponential growth, ignoring the messy reality of technical debt, talent warfare, and geopolitical friction. For instance, Chinese companies face export controls on advanced optics. The U.S. may restrict access to key laser chips. A supply chain disruption could shatter those profit forecasts overnight. Yet, there is a genuine opportunity here for the crypto ecosystem. DePIN projects like Filecoin, Helium, and Render need high-performance networking. As they scale to handle real-world workloads, they will face the same network bottlenecks that AI clusters face today. I expect a second-order effect: tokenized infrastructure will have to invest in faster optical interconnects, which will benefit the same hardware vendors. In that sense, Goldman’s bet on Zhongji is also a bet on the growth of compute-intensive applications, which includes crypto. Let me pivot to a contrarian angle. The market is currently obsessed with software narratives—AI agents, autonomous trading bots, zk-proofs. But the hardware layer is where the real constraints live. Software can be forked in a weekend. Hardware requires physical factories, supply chains, and years of testing. The optical module story is about the physical world imposing its own pace on the digital revolution. That’s a sobering thought for crypto maximalists who believe code can solve everything. It can’t. Bandwidth is physical. When I moved to Abu Dhabi in 2024, I facilitated roundtables between regulators and DAO founders. One thing that emerged was that even the most decentralized networks depend on centralized hardware vendors. The same companies that build AI infrastructure also build the switches and routers that power blockchain nodes. The narrative of decentralization often obscures this fact. If a single supplier—say Zhongji—controls 30% of the high-speed optical module market, that’s a point of centralization. A flaw, a black-box, a risk that the community must audit. So what’s the takeaway? I’m not calling for a sell-off of Zhongji stock. But I am saying that the narrative around “AI infrastructure = buy everything” is dangerously simplistic. The real alpha lies in understanding which bottlenecks will persist and which will be engineered away. Optical modules will remain a bottleneck for at least another five years. That makes Zhongji a case worth studying. But the profit growth Goldman predicts will only materialize if three things happen: (1) AI training continues to scale, (2) no major competitor emerges with a cheaper or better alternative, and (3) the supply chain remains geopolitically stable. Those are big ifs. Where capital flows, stories of value emerge. Right now, capital is flowing into photonics, lasers, and fiber. Crypto traders should watch this trend because it signals the health of the broader compute market. If optical module orders start slipping, AI and crypto token prices will follow. If they surge, the entire tech sector benefits. Decoding the noise to find the signal means not taking Goldman’s word at face value, but understanding the mechanics behind it. I’ll leave you with a question: In a world where every asset class is tokenized, who will manufacture the hardware tokens rely on? And will those manufacturers become the new whales of the digital economy? The sharding roots of tomorrow’s liquidity may start in a factory in Suzhou, not a smart contract on Ethereum.