We audit the code, but who audits the conscience? When Goldman Sachs raised its target price for Zhongji Innolight from 1,187 to 2,581 yuan, the market cheered. But I saw something else: a prophecy of centralization painted in the language of silicon photonics and scale-up networks. As an open source evangelist who spent years auditing the ethical soul of smart contracts, this report isn’t just about optical modules—it’s a mirror held up to the AI infrastructure stack that blockchain must now learn to read.
The Hook: A Price Target That Speaks Volumes
Goldman’s 118% jump in the target price for a Chinese optical module maker is not a random number. It’s a bet that the global AI bottleneck has shifted from GPU compute to interconnects. Over the past seven days, while the market fixated on NVIDIA’s earnings, a silent revolution occurred in the optics supply chain. Zhongji Innolight, an ODM/JDM supplier to giants like NVIDIA, Google, and Amazon, is now valued as if it owns the digital railroad of AI. The figure 2,581 yuan represents a belief that the next trillion dollars in AI capital expenditure will go not just into chips, but into the glass and silicon that bind them.
The Context: From Scale-Out to Scale-Up Networks
Let me translate the jargon. “Scale-out” networks connect many servers loosely—think of a library where each book is separate. “Scale-up” networks connect CPUs and GPUs tightly inside a single rack—think of a single brain with many neurons. As AI models grow, these internal connections become the bottleneck. The industry is migrating from 400G to 800G to 1.6T optical modules, each generation costing four to five times more. Goldman specifically highlighted silicon photonics—the integration of optics onto silicon chips using standard CMOS fabrication. This is not just a technical upgrade; it’s a paradigm shift.
Based on my audit experience analyzing DeFi protocols, I know that centralization often hides in the plumbing. Here, the plumbing is optical interconnects. The entire AI world is betting that Zhongji Innolight and a handful of others can scale their production lines fast enough to meet the demand from NVIDIA’s ultra-dense racks like the DGX GB200 NVL72. But who controls the glass? And more importantly, who controls the chips inside that glass?
The Core: A Technical Path to Centralization
Goldman’s thesis rests on three pillars: silicon photonics growth, scale-up network expansion, and higher-speed products. Each pillar carries a hidden cost for decentralization. First, silicon photonics lowers the cost of high-speed interconnects by leveraging existing semiconductor fabs. That sounds good, but it also means the entry barrier rises—only companies with access to advanced CMOS processes and deep packaging expertise can compete. In blockchain terms, this is like having only one node operator that can afford a full validator.
Second, the scale-up network forces AI compute clusters to be physically dense. A single rack of thirty-six NVIDIA B200 GPUs requires thousands of optical connections. This density creates a new kind of geography: the data center becomes a fortress of proprietary interconnects. No open standards. No permissionless participation. The very architecture of AI is being built with centralized hardware in mind.
Third, the shift to 1.6T and beyond means that the number of suppliers will shrink. Only those with the capital to invest in next-gen packaging and testing survive. The market is consolidating around Zhongji Innolight and Coherent. In a world where the decentralized web dreams of spreading compute across millions of edge devices, the AI backend is running in the opposite direction.
I dug into the numbers from my own research during the DeFi Summer. When I reverse-engineered the yield optimization of Harvest Finance, I found that the alpha came from unsustainable token emissions. Similarly, Goldman’s optimism assumes that AI capital expenditure will continue to grow exponentially. But what if it plateaus? What if the AI bubble bursts? Then the same optical module factories become stranded assets. And the decentralized networks that might have used that same infrastructure—like distributed compute marketplaces—are left without affordable components.
The Contrarian Angle: Where the Bottleneck Becomes a Beacon
Here’s where the contrarian in me wakes up. Goldman’s report is bullish on Zhongji Innolight because they see a monopoly in the making. But for the blockchain community, this concentration is a red flag. The same technology that powers centralized AI clusters could, if democratized, power decentralized physical infrastructure networks (DePIN). Silicon photonics, because it uses standard CMOS, could be fabbed in multiple foundries. The open source movement could fork the designs.
We saw this in the crypto space: when Ethereum’s gas costs became untenable, L2s like Arbitrum and Optimism forked the da layer and created their own scaling paths. Similarly, the optical interconnect standards (like OSFP, QSFP-DD) are open specifications. The lock-in is not in the standard, but in the proprietary firmware and licensing that vendors add. If the blockchain community starts building its own high-speed networking gear for decentralized compute clusters, we can bypass the vendor lock-in.
But the window is closing. The risk of export controls on high-speed optical modules (800G, 1.6T) is real. The U.S. could cut off China’s supply chain, or China could restrict exports. Either way, the global network of decentralized compute nodes becomes vulnerable. I saw this firsthand when I analyzed the custody solutions for Bitcoin ETFs in 2024—institutional adoption brought efficiency but also fragility. The same is happening with AI optics.
The Takeaway: Build Not for the Peak, but for the Plain
Goldman is building for the peak: the peak of AI hype, the peak of GPU demand, the peak of optical module prices. But we, the builders of decentralized systems, must build for the plain—the steady state where hardware is affordable, standards are open, and no single supplier holds the keys to the kingdom. The silicon photonics revolution is a double-edged sword. It can either deepen the centralization of AI infrastructure or become the backbone of a truly open compute fabric.
The question we should ask is not whether Zhongji Innolight’s stock will hit 2,581 yuan. The question is: who audits the conscience of the hardware vendors who are wiring the future of intelligence? We have open source audits for smart contracts. We need open source audits for optical interconnects. Not just for bugs, but for lock-in. For sovereignty.
Build not for the peak, but for the plain. That’s where the next generation of decentralized networks will grow.