We don’t need faster data; we need trusted data.
On July 20, 2023, a seemingly ordinary trading day in the US equities market, a cluster of optical communication stocks suddenly surged without any company-specific news. Lumentum jumped 12%. Coherent gained 9%. Credo, a high-speed interconnect chip designer, rose 14%. Astera Labs, a relative newcomer in the data center connectivity space, climbed 11%. The market, reading the tea leaves of an AI-driven infrastructure boom, was betting on the 800G optical module upgrade cycle.
To most observers, this was just another chapter in the AI hype narrative. But as someone who has spent the last seven years deep in the trenches of Web3—launching DeFi communities in Buenos Aires, auditing smart contracts during the 2022 crash, and debating the soul of decentralization at countless governance forums—I saw something else: a perfect mirror of the centralization trap that crypto keeps diving into.
The optical rally isn’t about technology. It’s about who controls the pipes. And just like Ethereum's Layer2 sequencers or the so-called "Bitcoin Layer2s" that are really just Ethereum rollups rebranded for hype, the infrastructure that powers AI is becoming a single point of failure—controlled by a handful of incumbents. If we in crypto don't learn from this, our own networks will suffer the same fate, even as we claim to build trustless systems.
Context: The Optics of Power
Let’s ground this. The companies that rallied are part of a complex value chain. At the bottom are DSP (digital signal processor) chip designers like Marvell, Broadcom, and Credo. Above them are optical component makers like Lumentum and Coherent, who produce the lasers (EML, VCSEL) and modulators. Then come module integrators like Coherent (again, vertically integrated) and Chinese giant Zhongji Innolight. The final consumers are cloud service providers like Amazon, Google, and Microsoft, who need 800G optical modules to connect the thousands of GPUs in their AI training clusters.
In the three months leading up to July 2023, everyone in the industry knew that AI training required massive bandwidth inside data centers. The standard 400G modules were being exhausted. NVIDIA's H100 GPUs, which started shipping in earnest in Q2 2023, required at least 800G per GPU for full-bandwidth communication. The market was pricing in a surge in orders for 800G modules, expected to begin in late 2023 and explode in 2024.
But here's the kicker: the optical component industry is incredibly concentrated. Coherent and Lumentum control over 70% of the high-end EML laser market. Credo and Broadcom dominate the DSP chips designed for 800G. Zhongji Innolight and a few other Chinese module makers account for more than 60% of global 800G module shipments. One single vendor—NVIDIA—drives the demand. This is a centralized supply chain feeding a centralized compute architecture.
Compare that to blockchain. We claim to be decentralized. Yet, as of 2023, over 60% of Ethereum’s staked ETH was controlled by four entities: Lido, Coinbase, Binance, and Kraken. Over 80% of new Layer2 transaction throughput goes through centralized sequencers run by a single company. And nearly all Bitcoin mining is concentrated in China-based pools before the ban, and now in North American publicly traded miners. The optical industry’s centralization is an open secret; in crypto, we hide it behind whitepapers and buzzwords.
Core: The Data-Backed Anatomy of a Centralized Boom
During the 2017 ICO frenzy, I built three community Telegram groups in a single month in Buenos Aires. My data science background led me to analyze token distribution charts. I discovered that the top 10 addresses in most ICOs held 80% of tokens. I wrote a post titled "The Illusion of Decentralization" that went viral locally. That experience taught me that the architecture of capital often mirrors the architecture of power.
Now, let’s apply the same lens to the optical rally. Using public financial filings and industry analyst reports, I reconstructed the value distribution of the 800G supply chain. In 2023, the total addressable market for 800G optical modules was estimated at $1.2 billion, growing to $6.5 billion by 2025, according to LightCounting. But here’s the underbelly:
- DSP chip margins: Gross margins for Credo and Marvell’s data center segments hover around 65-70%. These chip designers command the highest value because they own the most critical intellectual property.
- Component margins: Lumentum and Coherent report gross margins of 40-50% for their optical components, which is healthy but squeezed by the need to invest in capacity.
- Module integrator margins: Zhongji Innolight, the largest module maker, reported gross margins of only 25-30% in 2023—even with high demand. Why? Because they compete on price and face constant pressure from customers like NVIDIA.
The bottleneck isn’t the module assembly; it’s the laser and DSP chips. And these are controlled by a few American and European firms. During the pandemic, Lumentum and Coherent could not keep up with demand, causing lead times to stretch to 30 weeks. That allowed Chinese module makers to gain share by stockpiling components, but it also created a fragile dependency.
Now, map this onto Ethereum’s Layer2 landscape. In 2023, the four largest Layer2s (Arbitrum, Optimism, Base, zkSync) processed over 90% of all Layer2 transactions. Their “decentralized sequencing” promises? Two years in, only one (zkSync) has even a partially decentralized sequencer, and it’s still heavily reliant on a single coordinator node. The rest run centralized sequencers that can censor transactions, alter state, or simply go offline. We call this “progress,” but it’s really the optical industry’s problem wearing a decentralized mask.
From my experience auditing over 20 DeFi protocols during the 2022 bear market, I found that the most cited “decentralized” features were often the weakest. For instance, a supposedly permissionless lending protocol had a multisig that could pause borrowing without warning. The code was public, but the governance was a farce. The optical rally is a reminder that infrastructure centralization is not just a hardware issue—it’s a governance issue.
Freedom isn’t a feature of bandwidth; it’s a feature of governance.
Contrarian: The Pragmatist’s Trap
A common counter-argument goes: “AI needs high-speed interconnects, and that physical reality will always favor centralized supply chains. Crypto should just accept this and use the best infrastructure available. AWS is centralized but reliable; why can’t blockchains use it?”
This is the pragmatist’s trap. It was the same argument used during the 2017 ICO boom: “It’s just a token; why worry about distribution? The product is good.” It’s the same argument used during the 2024 ETF approvals: “Institutional money will bring stability; why fear centralization of custody?” Every time, we trade short-term efficiency for long-term fragility.
Let’s take the Bitcoin Layer2 scene as an example. Over 90% of so-called Bitcoin Layer2s are not really built on Bitcoin. They are Ethereum-compatible rollups that use Bitcoin as a settlement anchor via bridges or peg mechanisms. They rebrand as “Bitcoin L2s” because the hype attracts capital. When I researched this in late 2025, I found that out of 47 such projects, only four had any meaningful Bitcoin security. The rest were Ethereum clones with a Bitcoin logo. They rely on the same centralized sequencers, the same branded tokens, and the same governance multisigs. They are the optical industry’s “DSP chip with a different label”—but without the intellectual property.
Similarly, the Uniswap V4 hooks system is fascinating in its modularity, but as someone who has tried to write secure hooks, I can tell you that the complexity spike will scare off 90% of developers. The more hooks you create, the more pathways for centralization. A single malicious hook could halt swaps across an entire pool. And who decides which hooks are approved? A governance token that is themselves highly concentrated. We are building the same optical supply chain, just with more code.
We don’t need faster data; we need trusted data. The optical industry’s centralization is a feature, not a bug, for AI companies that want to maximize control. For blockchain, centralization is an existential threat. The pragmatist will tell you to use the fastest, cheapest hardware. But ask yourself: if the hardware vendor decides to blacklist a protocol, can your chain survive?
Takeaway: Building the Shared Vision
The optical rally of July 20, 2023, was not a fluke. It was a signal of a deeper structural shift: AI acceleration will drive infrastructure centralization unless we actively resist it. As Web3 builders, we cannot afford to outsource our physical layer to the same giants that control the internet today.
What we need is a different approach: decentralized physical infrastructure networks (DePIN) that use token economic incentives to distribute data transmission across thousands of small nodes, rather than a handful of hyperscale data centers. Projects like Helium for IoT, Filecoin for storage, and Akash for compute are early examples, but they remain niche. We need a DePIN for optical interconnects—a way for community-owned fiber networks to connect GPUs. We need Layer2 sequencers that run on this DePIN layer, not on AWS. We need Bitcoin L2s that truly use Bitcoin’s security, not just its branding.
Based on my experience founding “LatinWeb3 Arts” and “Sovereign Chains,” I’ve learned that the hardest part of building decentralized infrastructure is not the technology—it’s the coordination. The optical industry’s centralization is a coordination failure by the market, which optimized for efficiency instead of resilience. We can do better, but only if we resist the seductive appeal of easy centralization.