The math was sound; the trust was the variable. In the world of high-frequency transceivers, the math of AI network scaling has just been redefined by a single warning. B. Riley's recent note on network flattening is not just a call on optical modules; it is a systemic signal about the decoupling of capital expenditure from legacy infrastructure. As a macro watcher, I see this as a liquidity horizon shift, where the flows that once funded 400G and 800G gear are now being redirected into a new topology.
Context: The Global Liquidity Map for AI Infrastructure
Over the past 18 months, hyperscalers have spent billions on GPU clusters, each requiring a complex three-tier Clos network to interconnect tens of thousands of accelerators. This architecture, while proven, is inherently inefficient: it uses 10-20% of total system power just for switching, and introduces latency through multiple hops. The industry has been moving toward flattening—reducing the number of tiers by moving switching intelligence into the compute nodes themselves. B. Riley's warning argues that this flattening will crush demand for traditional low-rate transceivers (100G/400G) while accelerating the adoption of 800G and beyond.
But what does this mean for the global liquidity map? The hyperscalers' capex budgets are finite. If network flattening reduces the number of transceivers per rack by 30-40%, the total addressable market for optical modules may shrink even as speed increases. This creates a winner-take-lower dynamic where only the top-tier vendors with 1.6T+ capabilities survive. The capital that was flowing into diversified transceiver companies will now concentrate into a few key players, creating a liquidity vacuum elsewhere.
Core: Crypto as a Macro Asset in the Network Flattening Era
History does not repeat; it rhymes in code. The 2024 Bitcoin ETF cycle taught us that institutional capital flows are driven by infrastructure maturity. Now, the AI network flattening is a similar infrastructure shock—but one that could indirectly impact crypto mining and decentralized physical infrastructure networks (DePIN).
During the 2020 DeFi liquidity crisis, I built models predicting a 60% drawdown based on unsustainable yield mechanics. Today, I see a parallel: the demand for GPUs (used both for AI and crypto mining) is sensitive to the availability of high-speed networking. If traditional transceiver demand collapses, the supply chain for GPU interconnect (e.g., NVIDIA's NVLink, InfiniBand) could see delays or shortages, raising costs for miners. Conversely, DePIN projects like Helium or Filecoin that rely on traditional networking hardware may face slower deployments if new hardware becomes costlier due to the flattening shift.
I recently analyzed the machine-to-machine (M2M) transaction velocity in the context of AI agents—a framework I developed in 2026. The flattening of AI networks reduces the cost of data throughput, which could supercharge agent-based micro-economies on Layer 2 chains. However, this is a medium-term opportunity. The immediate risk is that the traditional networking incumbents (Ciena, Finisar, etc.) see a 20-30% revenue drop over the next 12 months, dragging down the broader semiconductor and tech market. Crypto, as a risk-on macro asset, tends to correlate with tech stocks—especially when the narrative shifts from growth to efficiency.
Correlation is the smoke; divergence is the fire. The B. Riley warning may cause a short-term dip in crypto due to risk-off sentiment, but the long-term divergence is clear: crypto's own network needs are different. Bitcoin mining is not dependent on high-speed optical interconnects; it relies on ASICs and relatively low-bandwidth communication. Ethereum's Layer 2 scaling also does not require 1.6T transceivers. Therefore, crypto may decouple from this specific tech sell-off. The real impact is on tokens that are marketed as "AI infrastructure" or "compute layer" projects—they may see their valuations adjust downward as the narrative shifts from "AI needs everything" to "AI needs less of the old stuff."
Contrarian: The Decoupling Thesis
The narrative dies when the ledger bleeds. But what if the warning is already priced in? I saw this pattern during the 2022 Terra collapse: the systemic risk was obvious only to those who understood the math of algorithmic stablecoins. Similarly, many in the networking industry have been expecting a flattening for years. The B. Riley note may simply accelerate the inevitable rotation from legacy to next-gen.
The contrarian angle is that the flattening will actually increase total transceiver demand in the long run—by enabling denser GPU clusters that require more fiber connections at the spine-leaf level. The average selling price per transceiver may drop, but volume could increase threefold as every GPU node gets multiple direct links. This is a scenario where the "crush on traditional transceiver demand" is only temporary, and the market rebounds with 1.6T volumes.
From a crypto perspective, this means that the hardware supply chain for mining may not be hurt as much as feared. Miners are moving to older nodes and cheaper chips, which are less sensitive to advanced networking. The real opportunity is in investing in companies that are enabling the flattening via silicon photonics or co-packaged optics—these are the equivalent of the 'infrastructure layer' tokens in crypto. I have been building a basket of such private companies, mirroring my 2024 ETF allocation strategy where I combined due diligence on custodial security with macro positioning.
Takeaway: Cycle Positioning Amidst the Flattening
We are watching the decay of leverage. The leverage in this case is the old triple-tier architecture. As it collapses, investors must reposition from legacy transceiver plays to the new interconnect standards. For crypto, the cycle positioning is clear: avoid AI-themed tokens that promise compute layer dominance without real hardware partnerships. Instead, focus on assets that benefit from lower-cost data throughput—DePIN projects, decentralized storage, and Layer 2s that can handle increased transaction volumes from AI agents.
Based on my 2017 ICO audit experience, I know that technological changes can create valuation bubbles before the infrastructure is ready. The flattening of AI networks is real, but the timeline for its full impact is 2-3 years. In the meantime, liquidity will shift to those who can bridge the old and new worlds. The math was sound; the trust was the variable. Trust in the network equipment makers who can pivot to 1.6T+ will determine who wins in the next cycle. For crypto, trust in the ability to maintain low-cost, secure networking will be the variable that separates the enduring protocols from the fragile ones.
