Last week, the US optical communication sector—Credo Technology, Astera Labs, Marvell, and Corning—surged more than 10% in a single session. Market chatter blamed AI hype. But for anyone who’s stared at enough whitepapers to know when a bottleneck is being priced, this wasn’t hype. It was a correction. The market finally realized that the next frontier of AI infrastructure isn’t faster GPUs—it’s the pipes that connect them.
I’ve been here before. Back in 2017, I audited over 40 Ethereum whitepapers during the ICO boom. I watched teams promise infinite scalability while ignoring the fundamental constraint: data bandwidth. That lesson re-emerges now, not in silicon but in blockchain’s own interconnect layer—the blob space that powers every Layer2.
Context: The Interconnect Age
Credo’s HiWire® Active Electrical Cables, Astera’s CXL Retimers, Marvell’s PAM4 DSPs, Corning’s Vascade® fiber—these aren’t household names. Yet they are the "toll roads" of the AI kingdom. When hyperscalers like AWS, Microsoft, and Google deploy 800G optical modules to link GPU clusters, every one of those modules needs a Credo DSP or an Astera Retimer. The market woke up to a simple truth: AI’s compute growth has outpaced its interconnect growth. The next phase of scaling will be defined by bandwidth, not teraflops.
Now transpose that to crypto. For the last two years, the narrative has been all about Layer2 execution: Arbitrum, Optimism, zkSync, Base. TPS numbers get thrown around like party confetti. But beneath that, a silent capacity crisis is brewing—the same kind of bottleneck that just lit up optical stocks. I’m talking about blob data availability. After the Dencun upgrade, Ethereum introduced blobs to give rollups cheap data space. But the space is finite. Each block can carry a maximum of 6 blobs (post EIP-4844), and the network is already approaching that ceiling during peak activity.
Core: When the Pipe Becomes the Constraint
Let me be direct: post-Dencun, blob data will be saturated within two years, and then all rollup gas fees will double again. This isn’t speculation—it’s a mathematical function of supply and demand. Every L2 transaction that posts data to Ethereum consumes a piece of blob space. As adoption grows linearly (and it will, especially with institutional DeFi and tokenized assets), the blob supply remains fixed until the next upgrade (PeerDAS, full Danksharding). We’re looking at a classic capacity cliff.
I saw this same pattern in 2020 when I founded OpenLedger Academy to teach yield farming. Back then, Ethereum’s base layer was the bottleneck. People focused on composability and TVL, ignoring that gas prices would soar as soon as DeFi summer arrived. Today, the focus is on execution, but the real constraint has moved up a layer. Blobs are now the scarce resource. And just like Credo and Astera are "sell picks and shovels" for AI interconnect, projects that provide data availability—Celestia, Avail, EigenDA—are the interconnect layer for crypto.
To understand why this matters, look at the technical analogy. In AI clusters, the bottleneck shifted from GPU compute to the SerDes and retimers that handle data transfer between accelerators. Similarly, in Ethereum, the bottleneck is shifting from L2 execution to the data channel that makes execution verifiable. Without sufficient blob capacity, rollups will either become expensive again or compromise on security by turning to off-chain committees. That’s not a scaling solution; it’s a trust regression.
During my 2021 soulbound NFT exhibition SoulBound Stories, I saw how digital scarcity creates identity. Scarcity of blob space creates something similar—a premium on data availability. Projects that can secure cheap, abundant DA will attract liquidity. Those that rely solely on Ethereum blobs will hit a cost wall. The market hasn’t priced this yet. Most traders still look at TVL and TPS. They ignore the cost per byte of data posted.
Contrarian: The Misplaced Faith in "Infinite Scalability"
Here’s the contrarian angle: every crypto investor I meet believes Danksharding will solve everything. "Once full sharding comes, blobs will be infinite." That’s the same line I heard in 2017 about Plasma and state channels. In practice, full Danksharding is years away (likely 2026–2027), and even then, the scalability of blobs isn’t infinite—it’s bounded by node hardware constraints. Meanwhile, demand from real-world adoption (RWA tokenization, cross-chain bridges, AI oracle feeds) will outstrip supply long before the upgrade lands.
This creates a window for modular data availability solutions. Celestia, Avail, and Near’s EigenDA are building separate DA layers that can scale independently. They face their own challenges—weak economic security, dependence on an honest minority—but they offer a pragmatic escape hatch. In the same way, Credo’s AEC cables solve the copper-versus-fiber dilemma for short-reach AI connectivity, modular DA layers solve the "blob or bust" dilemma for L2s. Don’t bet against pragmatic off-ramps just because they aren’t "pure Ethereum." The market will choose whatever keeps fees low.
I saw this in 2021 when I curated SoulBound Stories—artists chose platforms based on community alignment, not technical perfection. Users will follow what works. If Ethereum blobs become too expensive, L2s will migrate to Celestia or EigenDA, fragmenting liquidity but unlocking lower costs. That fragmentation is painful for composability but inevitable. It mirrors the AI world where hyperscalers use a mix of optical, AEC, and copper depending on distance and latency.
Takeaway: Redefining "Scaling"
We need to stop measuring scaling by TPS. The real metric is data throughput per dollar. The next bull run won’t be led by the fastest L2—it will be led by the one with the most efficient data pipe. Ethereum’s blob space is the optical fiber of crypto. And just like Corning, Credo, and Astera are now market darlings, the projects that provide that pipe—Celestia, Avail, and even Ethereum itself when Danksharding arrives—will be the compounders of this cycle.
Democracy isn’t a transaction where every voice holds weight. Code is the new conscience. And the only way to scale democracy, or code, is to first scale the channel through which truth flows. Pay attention to the pipe. The pipe is the product.