Zhongji Innolight filed for a $7 billion Hong Kong IPO. That’s more than the market cap of most Layer 1 blockchains. It’s also a stark reminder that the next bottleneck in digital assets isn’t code—it’s copper, glass, and silicon.
This isn’t a crypto company. Zhongji Innolight makes optical modules—the tiny transceivers that connect GPUs in AI data centers. Its customers are Nvidia, Microsoft, Google. The IPO proceeds will fund factories for 1.6T modules and silicon photonics. While crypto circles obsess over zkEVMs and parallel EVMs, the real infrastructure game is being played in Shenzhen and Santa Clara.
Context: The Hardware That Powers Both AI and Crypto
Zhongji Innolight is the world’s largest supplier of 800G optical modules. These are the physical links that let thousands of Nvidia H100s talk to each other. Without them, the largest AI clusters would grind to a halt. The same hardware is used in crypto mining farms—especially for ASIC-based networks like Bitcoin, but also for GPU-based coins like Ethereum Classic or projects on filecoin that require high-bandwidth storage networks.
The IPO is a direct consequence of the AI arms race. OpenAI, Google, and Meta are ordering clusters with 100,000+ GPUs. Each cluster needs millions of optical links. Zhongji, along with Coherent and Innolight (a separate entity), controls over 70% of the high-speed module market. The $7 billion raise is for capacity expansion—building lines to produce 1.6T and eventually 3.2T modules. This is a bet that AI compute demand will continue to double every year.
But here’s the rub for crypto: the same supply chain that serves AI also serves decentralized compute networks. Every Filecoin storage provider, every Akash compute node, every Ethereum validator that relies on cloud infrastructure is downstream of these same optical module factories. When a factory in Wuhan shuts down for COVID, both GPT-5 and the Filecoin network feel the delay.
Core: Systematic Teardown—Three Vulnerabilities Crypto Cannot Ignore
1. Supply Chain Concentration
Zhongji Innolight and its primary competitor, Coherent (formerly II-VI), together control roughly 75% of the 800G module market. That’s a duopoly. The top three module buyers—Nvidia, Microsoft, and Google—account for an estimated 60% of revenue. This concentration creates a single point of failure: if a trade war escalates or a factory experiences a fire (as actually happened in 2020 with a Nippon Electric glass plant), every downstream user suffers.
Crypto projects that market themselves as decentralized—Akash, Livepeer, Arweave—are exposed to this risk. Their nodes run on hardware that depends on a handful of suppliers. On-chain data from Dune Analytics shows that over 60% of Ethereum validators use centralized cloud providers like AWS or Hetzner. Those providers are the ones buying Zhongji’s modules. The tech stack may be open-source, but the physical layer has become a trust-based network.
2. Geopolitical Friction
Zhongji Innolight is a Chinese company. Its listing in Hong Kong, while technically outside mainland jurisdiction, places it squarely in the U.S.-China tech war. The Biden administration’s export controls on advanced chips already restrict Nvidia from selling its highest-end GPUs to China. Optical modules are not yet sanctioned, but the logic is similar: if AI compute is strategic, so is the connectivity that makes it possible.
From my audits of DeFi protocols, I’ve seen how quickly regulatory risk can spread down the stack. When Tornado Cash was sanctioned, it wasn’t just the front-end that broke—it was the infrastructure providers, the RPC endpoints, the GitHub repositories. A similar cascade could happen here. If the U.S. restricts the export of optical module DSPs (digital signal processors) to China, Zhongji’s production slows. That directly impacts Nvidia’s cluster delivery timelines. And since crypto mining and AI share the same hardware supply, the ripple effects would hit GPU prices for miners and stakers.
3. The Centralization Paradox
Crypto’s core narrative is decentralization. But infrastructure reality is centralization. The largest machine learning models run on a handful of massive clusters. The largest blockchain networks run on a handful of cloud providers. Zhongji’s IPO highlights this paradox: the hardware that enables AI’s scale also enables crypto’s transactional throughput. But that hardware is built by a small number of capital-intensive firms.
Consider the Filecoin network. It stores over 2,000 PiB of data, much of it on GPUs and SSDs. But the connectivity between storage providers and retrieval markets relies on optical networks. If the cost of 800G modules stays high due to duopoly pricing, Filecoin storage providers face higher operating costs. That reduces margins and consolidates power among large-scale miners who can afford the hardware. Decentralization becomes a myth.
Data signals from the IPO filing
We don’t have the final prospectus yet, but industry estimates suggest Zhongji’s revenue grew at a CAGR of 40% over the past three years, driven by AI demand. Gross margins hover around 35%—healthy but under pressure as competition heats up. The $7 billion will likely be split: 40% for new production facilities, 30% for R&D (silicon photonics, CPO), 20% for working capital, 10% for potential acquisitions.
First-person technical experience: Auditing infrastructure dependencies
In my work auditing crypto security, the most common vulnerability isn’t a smart contract bug—it’s single points of failure in off-chain infrastructure. I’ve seen staking pools that rely on a single VPS provider, bridges that use a single oracle node, and DeFi protocols whose admin keys are stored on a single cloud server. Zhongji’s IPO is a warning for the industry: we must extend our audit scope to include the physical supply chain. Are the optical modules in your validator’s data center coming from a single vendor? Are you diversified across manufacturers and regions?
Contrarian Angle: What the Bulls Get Right
It’s easy to be cynical. But the IPO could also be a catalyst for positive change. First, capital markets discipline brings transparency. Zhongji will now have to disclose revenue concentration, customer dependencies, and R&D spending quarterly. That data can be used by the crypto ecosystem to model risk. Second, capacity expansion will eventually drive down optical module costs. The price per Gbps has been falling historically—from $10/Gbps in 2018 to under $1/Gbps today for 800G. With more competition, the cost of interconnecting decentralized compute nodes could drop significantly.
Third, the IPO legitimizes hardware as an asset class. This could attract institutional capital to physical infrastructure funds that invest in data centers and networking equipment. Some of that capital might flow into decentralized infrastructure projects. For example, if Zhongji’s new 1.6T modules are open-standard and compatible with emerging decentralized storage networks, it could accelerate network capacity. The bulls argue that centralization at the hardware layer is a necessary evil to achieve scale, and that once the scale is reached, democratization happens.
But I’ve seen this play before. In 2017, BitConnect had a whitepaper that promised decentralized lending. In reality, it was a centralized ponzi with no code. In 2022, we learned that the entire Solana ecosystem was dependent on a single RPC provider for months. The pattern is clear: every miracle protocol has a centralized hardware dependency hiding in plain sight. Zhongji’s IPO is not a bug—it’s a feature of the current economic system. The question is whether crypto will build redundant, resilient hardware layers or continue to rely on monocultures.
Takeaway: The Accountability Call
The blockchain industry loves to talk about “trustless” systems. But “trustless” only works if the hardware layer is trust-minimized. Zhongji’s $7 billion IPO proves that the real trust is placed in a few factories in Shenzhen and a few design firms in California. Until the crypto ecosystem invests in its own physical infrastructure—decentralized manufacturing, open-source optical modules, community-owned data centers—we’re just building sandcastles on a centralized foundation.
AI models are black boxes until you audit the hardware supply chain. The whitepaper is fiction; the supply chain is fact. Every miracle protocol has a centralized hardware dependency hiding in plain sight. Zhongji’s IPO is the smoke. The fire is yet to come.