SN51's Lium has quietly burned close to $2.7M in alpha over six months, funded entirely by GPU rental billings that just posted a 36% monthly jump. Elsewhere, the Yuma Total Market Fund matched TAO's best month yet, SN33's ReadyAI is closing in on six-figure ARR, and SN91's Cascade model claims a training-efficiency win worth watching.
Full coverage
4 reports
01 / QUICK HIT
Lium runs a decentralized GPU marketplace on SN51, paying miners 95% of rental revenue and using the rest to buy and burn SN51 alpha. The mechanism has now run five rounds between April and September, pulling roughly 171,000 alpha tokens out of circulation for nearly $2.7M total. Growth is compounding: GPU billings hit $964K over the 30 days to September 9, a 36% jump from the prior period, with billed accounts nearly tripling from 445 to 1,187.
September buyback: $946K — more than double August's $365K
~613,000 GPU-hours rented in the latest 30-day window
171,000 SN51 alpha burned since April across five buybacks
02 / DAILY
Yuma Fund Posts Its Best Month
The Yuma Total Market Fund returned 30.3% in September on a USD basis — its strongest month since launch and roughly in line with TAO's own 30.9% gain. Over its first full quarter, the fund edged out TAO outright, returning 49.0% versus TAO's 48.0%. The fund blends bittensor-native assets with subnet tokens, so the parity with TAO's move suggests broad-based subnet strength rather than a single outlier.
03 / DAILY
ReadyAI Nears $100K ARR
SN33's ReadyAI is approaching six-figure annualized revenue from its Jobs API, with the latest run-rate at $95K ARR. The subnet's September alpha buyback came in 45% higher than August's, tracking the same revenue-to-buyback pattern showing up elsewhere in the ecosystem this cycle.
04 / DAILY
SN91's Cascade Claims an Efficiency Edge
SN91's Cascade model reportedly beat the Toto 2.0 4M model on both TIME MASE and TIME CRPS benchmarks while training on just 15.4% as many tokens. The training run used only synthetic, miner-generated data over 58 days — a claim worth tracking for independent verification given the small data footprint and the self-generated training set.
