Spent more time thinking about what comes after the AI chip trade, and I think the market may be staring at an even bigger bottleneck: DATA

Not raw internet sludge or another database with an AI label slapped on it.
I mean clean, licensed, attributable and increasingly real-world data that models can legally train on and actually learn something new from.
Compute is still important, but it’s no longer the only scarce input.
Morgan Stanley sees around $2.9T to $3T of AI infrastructure spending through 2028, with less than 20% deployed so far.
There will be more #GPUs, more data centers, more power contracts and more inference efficiency.
What doesn’t automatically scale with them is fresh human knowledge and physical-world experience.
Physical AI data is even more expensive, embodiment-specific, hard to synthesize perfectly and impossible to scrape at sufficient depth.
The bottleneck is capturing the right event, with the right sensors, under the right rights, and proving where it came from.
Private markets already seem to understand this before public markets do.
– Scale AI was valued at nearly $29B after Meta’s $14.3B deal for a 49% stake.
– Surge AI did more than $1B in revenue in 2024 while seeking a $15B valuation.
So if a major catalyst is coming in TradFi, crypto may also have its leg in AI data collection and provenance.
– @eigencloud | $Eigen: EigenDA powers verifiable data availability, while EigenCompute, EigenAI and EigenVerify let agents prove where data came from and how outputs were generated.
– @grass | $Grass: turning idle bandwidth into verifiable AI training data. 2.5M+ nodes across 190+ countries scrape ~100TB/day, with ZK proofs making every dataset traceable and auditable.
– @vana | $Vana: attacking AI's data bottleneck by turning user data into permissioned, provable training datasets. 1M+ contributors, DataDAOs and portable AI memory all live on one network.
– @SaharaAI | $Sahara: AI-native L1 where contributors collect, label and own datasets while provenance, licensing and royalties stay onchain. Trying to make high-quality AI data scalable.
– @datafdn | $Data: provenance, consent, licensing and audit rails onchain, while the Poseidon layer cleans and labels real-world data for model training.
– @oceanprotocol | $Ocean: turning data into onchain assets with Data NFTs, Compute-to-Data and provenance rails so AI can train on private datasets without exposing the raw data.
– @origin_trail | $Trac: shipping DKG V10 where AI agents share verifiable memory. dRAG, provenance and multi-agent memory make data reusable without losing trust.
– @datainetwork: turning raw blockchain logs into structured, machine-readable intelligence. 3.5B+ txns indexed, 2.5M contracts labeled and 150M+ DeFi events for AI agents to reason on.
Not a call to ape every ticker above. Just a map of who's building the infra before the market prices them.