A former ByteDance employee turned investor made 30 million yuan by betting on AI storage stocks. The story spread through crypto circles like wildfire. Not because it was about a successful trade—but because it exposed a truth many in Web3 prefer to ignore: the infrastructure race is being won by centralized giants, while decentralized alternatives still struggle for adoption.
I read the report carefully. Leto Bao, a 28-year-old ex-ByteDancer, identified a price anomaly on Pinduoduo for hard drives in mid-2023. He dug deeper, realized data centers were hoarding storage for AI workloads, and quietly bought shares of Micron, SK Hynix, and Samsung. Eight months later, his portfolio grew by 30 million yuan. He quit his job. The narrative was perfect: a self-taught analyst outsmarted the market by reading supply chain signals.
But what if the same signals point toward a different kind of storage—one that aligns with the principles of decentralization? The Web3 community often dismisses centralized storage as a relic. Yet the market tells a different story. Traditional storage companies have collectively added over $200 billion in market cap since ChatGPT launched. Decentralized storage tokens like Filecoin, Arweave, and Storj have underperformed relative to their hype. Why?
The answer lies not in the technology, but in the economic incentives.
Centralized storage providers—AWS, Google Cloud, and the hardware makers—have a clear, unified business model. They offer fast, reliable, low-latency storage at scale. Their customers are AI labs and enterprises that prioritize speed and compliance over censorship resistance. Data sovereignty is a luxury, not a necessity, for most AI training pipelines. When ByteDance or OpenAI needs to store petabytes of training data, they don't ask about tokenomics. They ask about IOPS and uptime SLAs.
Decentralized storage, on the other hand, is built on a different premise: that data should be owned by users, not corporations. Protocols like Filecoin use proof-of-replication and proof-of-spacetime to ensure data is stored physically across many nodes. Arweave offers permanent storage with a one-time fee. These are remarkable technical achievements—but they come with trade-offs. Retrieval latency is higher. Costs can be unpredictable. And for a large AI training cluster, the need for rapid data access often overrides ideological purity.
Leto Bao's story is a wake-up call. It shows that the most reliable way to profit from the AI revolution, for now, is to bet on the plumbing—the hardware that makes AI possible. But the plumbing itself is evolving. The next wave may not be centralized.
Consider this: the same scalability bottlenecks that plague blockchains are now hitting AI storage. Centralized data centers are running out of physical space, energy, and cooling capacity. The rise of long-context models (like GPT-4's 1M+ token context) demands exponentially more memory and storage. Companies are exploring disaggregated storage architectures, where storage and compute are separated and connected via high-speed fabrics like CXL. This architectural shift opens a window for decentralized solutions.
Enter the Decentralized Storage Trilemma.
Every storage network—whether centralized or decentralized—faces a trilemma between speed, cost, and trust. Centralized systems optimize for speed and cost, sacrificing trust. Decentralized systems optimize for trust and cost (via commoditized hardware), sacrificing speed. But AI workloads need all three. That's the gap.
Protocols like Filecoin are bridging it by introducing retrieval markets and content delivery networks (CDNs) that cache hot data closer to users. Arweave's bundling services allow cheaper writes by aggregating transactions. Storj's edge network reduces latency. These are incremental improvements, but they signal a direction: the future of AI storage may be hybrid—centralized for hot data, decentralized for cold and warm archival.
Leto Bao's success came from analyzing real-world demand signals. He saw the spike in hard drive prices and understood that it was not a temporary glitch but a structural shift. If we apply the same lens to decentralized storage, what do we see?
On-chain metrics tell a quiet story.
Filecoin's total storage capacity has grown from 1 EiB in 2021 to over 20 EiB in 2025. Active deals—contracts where clients pay FIL to store data—have increased 5x year-over-year. Arweave's permaweb now holds over 150 TB of data, including archives of scientific papers, DAO treasuries, and NFT metadata. Storj's network has processed over 10 billion files.
These numbers pale in comparison to AWS S3 (over 100 exabytes of stored objects), but they are growing from a sincere base. More importantly, the client composition is shifting. A growing share of storage deals come from AI projects: researchers storing model versions, decentralized compute networks caching training data, and DAOs archiving governance records.
Yet the contrarian truth is this: decentralized storage is not yet ready for prime-time AI training. The latency and cost barriers remain significant. An investor who bought Filecoin in 2021 hoping to ride the AI wave would have lost 80% of their capital. Meanwhile, the centralized storage investor—the one who bought Micron or Seagate—would have doubled or tripled their money.
This is the uncomfortable reality that the crypto community often avoids. We evangelize decentralization as a moral imperative, but the market rewards pragmatism. The ByteDance investor didn't care about blockchain. He cared about hard drives. That's why he won.
So what does this mean for the decentralized storage movement?
It means we need to stop treating decentralization as a panacea and start treating it as a niche solution with specific use cases. For AI archival, disaster recovery, and data sovereignty-sensitive applications (like medical records or government documents), decentralized storage offers unique value that centralized systems cannot match. For high-frequency AI training, it remains a second choice.
But the momentum is building. As AI models become more diverse and training data becomes more personal (e.g., fine-tuning on user-generated content), the demand for user-controlled storage will rise. The EU's Data Act and similar regulations are pushing for data portability and user ownership. These macro trends favor protocols that give individuals control.
From the ashes of the 2022 bear market, we planted seeds for 2030.
Leto Bao's story is a reminder that early adopters of any infrastructure wave profit immensely. The first wave was centralized storage for AI. The second wave may be decentralized storage for AI. The investor who buys Filecoin, Arweave, or Storj today—and holds through the volatility—might be the next 30 million yuan story.
But only if they understand the technology well enough to distinguish real demand from speculative hype. The ByteDance investor studied the supply chain. The Web3 investor must study the protocol economics.
The real question is not whether decentralized storage will win. It is whether it will win in time for the next AI scaling cycle.
In the next three years, the volume of data generated by AI will exceed all previous human-created data combined. If decentralized storage can capture even 1% of that market at a 10x premium for data sovereignty, the token valuations will be staggering. But if the technology fails to scale, the entire sector risks being relegated to a boutique niche.
I have spent the last five years building communities around decentralized technologies. I have seen too many projects promise the moon and deliver a pebble. But I have also seen the quiet, steady work of engineers solving latency problems, designing better proof systems, and integrating with existing cloud infrastructure. The foundations are being laid.
Silence is the sound of true development.
The ByteDance investor listened to the market. The next great Web3 alpha will come from listening to the network. Watch the storage deals, watch the retrieval fees, watch the partnerships. The signals are there.
As of mid-2025, the decentralized storage market cap sits at roughly $15 billion—a fraction of the $200+ billion centralized storage market. But every exponential technology starts small. The question is not where we are today, but where we will be when the next wave of AI demand hits.
Will the next millionaire be a Filecoin miner in Manila, or a Micron shareholder in New York?
The answer depends on which infrastructure we choose to build.
From the ashes of the centralized storage boom, we are planting the seeds of a decentralized future. But seeds need time, water, and patience. The investor who understands that—and acts on it—will be the one who writes the next story.