$11.49M ARR now buys PYTH, with $2.09T RWA volume behind it ⚡
$PYTH is sitting under the markets where stocks, gold, oil and index perps already trade at serious size. $SOL has the fast apps, liquidity and high-throughput audience that understands why speed needs reliable data underneath.
Pyth closed Q3 at $11.49M ARR, up 86% from Q2.
The DAO now sends 100% of the funds it receives from Pyth products into open-market PYTH purchases.
Now add the usage number.
Q3 RWA perps did $2.09T in volume, and 94.1% of it ran on venues powered by Pyth.
That is not a small footprint.
When traders touched RWA perps last quarter, a large part of that market was already leaning on Pyth data.
Kalshi uses Pyth prices for stock and commodity markets. Polymarket uses Pyth for real-world asset markets. Coinbase uses Pyth data for continuously priced thematic markets.
So $PYTH now has the three things markets usually want to see together:
Usage. Revenue. A clearer route from product receipts to token buying.
$ARB is where Bankr is testing its newest launch mechanics, and the most revealing decision is one almost nobody covered. $UNI shipped v4 hooks, and Bankr chose not to depend on them for these launches.
Arbitrum launches moved to v3 pools instead.
The reasoning is practical rather than ideological. A v3 pool is routable by aggregators the moment it exists. A pool wrapped in a custom hook may not be, which means a token can launch into a market that most interfaces cannot quote. Liquidity that cannot be routed to is liquidity in name only.
They also pointed at 0x's writeup on hook safety as a reason not to build a launch path that depends on them.
That is the part worth generalising beyond this one product.
Hooks are genuinely powerful and they are also new surface area, both for exploits and for integration gaps. A launchpad's job is to produce a market people can trade in on day one, which argues for the boring pool and the custom logic sitting elsewhere.
Bankr put the new fee mechanics in the contract rather than in a hook, which gets most of the flexibility without the routing cost.
The counterargument is real. Hooks are where the interesting design space is, and anyone avoiding them trades capability for compatibility.
My read is that compatibility wins at launch and capability wins later. Shipping a token nobody can route to is a worse failure than shipping one with ordinary mechanics.
I think the AI trade is starting to move from models toward the infrastructure that keeps those models running. $RENDER is a good example of that shift.
Render’s Compute Client stack now opens its distributed GPU network to training, inference and fine-tuning, while it is separately onboarding data centers with 25+ GPUs or H100-class hardware and above.
Bittensor is moving in a similar direction through a completely different model.
Across TAO, subnets can compete to produce digital commodities such as compute and inference, while newer subnet infrastructure is already being built around B200/B300 and RTX Pro 6000 capacity.
That is what makes $B3 interesting to me.
B3IQ takes a third approach: a customer-owned neocloud.
Instead of only aggregating GPUs or incentivizing operators, users can own the actual bare-metal machine while B3IQ handles procurement, hosting and operations. Spare capacity can then be made available to outside demand when the owner chooses.
Three different models are emerging:
Render aggregates compute.
TAO incentivizes it.
B3IQ makes the cloud itself customer-owned.
If AI demand keeps compounding, I think the real battle becomes who can coordinate hardware, capital and utilization most efficiently.