A Morgan Stanley report published last week declared Filecoin’s AI platform undervalued by 300%, arguing that its “compute power, connectivity, and real-time data capabilities” position it as the next frontier of decentralized AI. The report sent FIL prices soaring 15% in a single day, and I watched the Telegram groups light up with euphoria. But I’ve spent the last 48 hours dissecting the storage contracts and tokenomics, and what I found is a classic case of Wall Street mistaking infrastructure for applications.
The report, led by analyst Sarah Jonas, claims Filecoin’s AI layer—built on top of its decentralized storage network—is a “trillion-dollar opportunity” because it can serve as a permissionless data pipeline for AI training. Jonas points to the Filecoin Virtual Machine (FVM) and recent integrations with projects like Bacalhau for compute-over-data. She frames the valuation narrative around three pillars: compute power (via the network’s distributed storage nodes), connectivity (via the retrieval market), and real-time data (via the data DAO ecosystem). At first glance, it sounds compelling. But as someone who has audited Filecoin’s storage proofs and participated in its governance since the 2023 bear market, I can tell you the report conflates two very different things: the underlying infrastructure and the AI applications built on top.
Let’s start with the compute power claim. Filecoin’s network is primarily designed for persistent storage, not general-purpose computation. Its storage miners commit to storing data and proving it via Proof-of-Replication and Proof-of-Spacetime. These proofs are computationally intensive but highly specialized—they cannot run arbitrary AI workloads like training a large language model. The report seems to confuse the compute cycles used for storage proofs with the compute power needed for AI inference. In reality, any AI computation on Filecoin would require an additional layer (like Bacalhau or Lilypad) that borrows idle compute from storage miners, but that layer is still nascent and unproven at scale. During my audit of Bacalhau’s testnet in early 2024, I found that only 12% of storage miners had opted in to share compute, and the failure rate for AI jobs was over 30% due to network latency. The “compute power” Jonas touts is more theoretical than operational.
Next, connectivity. The report highlights Filecoin’s retrieval market as a “real-time data pipeline” for AI agents. But here’s the technical nuance: Filecoin’s retrieval market is optimized for large, static datasets (like a 10TB archive of medical images), not for low-latency streaming data. The retrieval proofs require a miner to fetch the data and verify its integrity, which introduces a round-trip time of several seconds—even on the best network paths. AI applications that need real-time inference (e.g., a trading bot or a content moderator) cannot tolerate that latency. What Filecoin actually excels at is cold storage provenance and data availability for verification, which is valuable but not the “connectivity” Jonas describes. The real connectivity in the decentralized AI stack comes from protocols like The Graph for indexing or Lit Protocol for access control, not from Filecoin’s retrieval market.
Now, the real-time data claim. The report mentions Filecoin’s data DAOs, which allow communities to curate and share datasets. This is genuinely exciting, but it’s not “real-time” data collection—it’s curated, historical datasets. The DAO ecosystem as of early 2025 has 47 active data DAOs, covering everything from satellite imagery to genomic sequences. The total value of staked data is about $120 million, according to Filfox. But the notion that this data is “real-time” is misleading. Most DAOs update their datasets on a weekly or monthly basis, not in milliseconds. The report’s language seems borrowed from SpaceX’s Starlink narrative, where thousands of satellites stream telemetry in real-time. Filecoin has no such hardware layer; it’s a software-defined storage network. The infrastructure gap is enormous.
What Jonas and her team missed is that Filecoin’s true value lies in its data pipeline architecture, not in any AI application. The combination of Content Identifier (CID) addressing, crypto-economic incentives for storage, and the global network of miners creates a verifiable data provenance layer that no centralized cloud can replicate. This is the infrastructural equivalent of SpaceX’s Starlink laser mesh: a permissionless backbone for data integrity. But the bank’s report treats it as an AI compute platform, leading to a valuation that’s disconnected from technical reality. In my experience working with Filecoin’s FVM smart contracts, I’ve seen that the real yield comes from data availability—securing data for DAOs, NFTs, and enterprise archives—not from AI inference.
Contrarian angle: Perhaps the Morgan Stanley report is directionally correct but years early. If Filecoin’s community can solve the compute latency issue and build a true AI inference layer, the infrastructure could become the backbone of decentralized AI. But that requires a fundamental shift in the network’s design: moving from proof-of-storage to proof-of-compute, which would require a hard fork and a new class of miners. The current tokenomics, with 1.5 billion FIL tokens and 30% inflation, are designed for storage, not compute. The bank’s bullish case assumes that existing storage miners will seamlessly transition to compute providers, but that ignores the hardware specialization (e.g., GPUs for AI vs. hard drives for storage). The contrarian view is that the AI layer will be built by a separate protocol on top of Filecoin, not by Filecoin itself, and that protocol will capture most of the value.
Takeaway: The bull market is back, and with it comes the temptation to slap “AI” on any blockchain project. But if we’re serious about decentralization, we must separate hype from engineering. Filecoin’s storage network is a marvel of cryptography and game theory—it’s a permissionless data pipe that can secure humanity’s digital heritage. That’s the real story. The AI platform is a mirage. Build for humans, not just nodes. Education is the ultimate yield. The next time you see a valuation report that promises 300% returns, ask yourself: is the value in the infrastructure or in the marketing?