On a seemingly ordinary Tuesday, cloud stocks like CoreWeave and Nebius jumped alongside memory giants SK Hynix, SanDisk, and Western Digital. The headlines screamed "AI-driven rally" and "storage cycle recovery." But as someone who has spent years watching centralized choke points form in both traditional finance and tech, this price action tells a different story—one of growing concentration that blockchain was supposed to prevent.
The rally celebrated a system where the most valuable components of AI—compute and memory—are controlled by a handful of players. SK Hynix dominates HBM (High Bandwidth Memory) with over 50% market share, while Samsung and Micron follow. The cloud providers like CoreWeave are essentially NVIDIA's downstream proxies. This is not the open, permissionless future we envisioned. It's a walled garden with titanium fences.
Connect first, transact second. Always.
I remember 2016, when I wrote my first Spanish-language tutorial on trustless collaboration. Back then, the promise was simple: decentralized infrastructure would remove single points of failure and democratize access. Fast forward to 2025, and the AI boom is funneling capital into the most centralized parts of tech. The protocols we built—Filecoin for storage, Akash for compute—are still niche. Meanwhile, the market celebrates CoreWeave, a company that, despite its "cloud" label, is as centralized as it gets: it leases GPU clusters from a single supplier (NVIDIA) and sells them to enterprises. No redundancy, no community governance, no transparency.
The memory chip side is even more revealing. HBM is the lifeblood of AI training, and its supply chain is a poster child for concentration risk. Three Korean firms control over 90% of the market. In my work evaluating DeFi protocols, I often criticize Aave's and Compound's arbitrary interest rates—disconnected from real market supply and demand. But at least those rates are on-chain and auditable. Try getting a transparent cost breakdown for HBM. The pricing is opaque, tied to multi-year contracts with NDAs, and influenced by geopolitical pressures. Does that sound familiar? It should—it's the same problem Tether has with its reserves: we pretend the opacity doesn't exist because the system works for now.
The market is ignoring the sword of Damocles hanging over this rally.
Post-Dencun, we've seen how quickly blob data can saturate Layer 2 throughput. The same will happen to HBM capacity if AI demand keeps growing at 30-50% CAGR. The memory giants are building new fabs with billions in capex, but those take 2-3 years to ramp. And what happens when Samsung catches up with HBM4 and starts a price war? SK Hynix's entire valuation depends on being the sole supplier to NVIDIA—a single point of failure that rivals the risk of a Tether blacklist.
The contrarian angle here is that the AI rally is actually making the case for decentralization stronger, not weaker. Every time we celebrate a centralized cloud stock, we should ask: how much of that value could be captured by a decentralized equivalent? Akash's compute marketplace offers GPU leasing without a middleman. Filecoin's storage network ensures data is distributed across thousands of nodes, not held in a single Amazon S3 bucket. Yes, they lack the raw performance of HBM-backed clusters. But for inference tasks, for smaller AI workloads, for privacy-sensitive applications, decentralized alternatives are already viable. The problem is narrative capture—everyone is looking at the stock market instead of the protocols.
Based on my experience auditing decentralized protocols, I've learned that fragility is always priced in after the crash, never before.
When Terra collapsed in 2022, everyone blamed the code. But the real issue was a single point of failure—the Luna Foundation Guard's concentrated BTC reserves. Today, the AI supply chain is similarly concentrated. If a trade war cuts off HBM supply to China, or if NVIDIA decides to vertically integrate memory production, the entire ecosystem trembles. Decentralized networks don't have that vulnerability because they're built on commodity hardware and open markets.
So what does this mean for us? It means the next bull run won't be about price alone. It will be about infrastructure resilience. The protocols that survive will be those that offer a genuine alternative to centralized choke points—not by competing on peak performance, but by being permissionless, auditable, and antfragile.
The question I leave you with: Are you betting on the walled gardens, or on the open plains?
The memory and cloud rally may continue for months. But the cracks are already there. Those who invest in decentralized physical infrastructure networks (DePIN) now may find themselves the quiet winners when the next concentration crisis hits. Build the alternative, not the backup.