0x275f80d8275f…275f80d5
Bittensor's Lium Subnet Hits $34.6M Annualized GPU Spend Run Rate
Lium, Bittensor's Subnet 51, booked $2.70 million in GPU rentals across 31 days and 2,394 customers. Its $34.6 million figure is a seven-day extrapolation, not collected revenue.

Outputs
Lium recorded $2.70 million in GPU rentals over 31 days (Oct. 10–Nov. 9) with 2,394 billed customers.
The $34.6 million annualized figure extrapolates only the latest seven days of spending, roughly $95,000 per day.
Monthly spending rose 87% from $1.45 million; billed customers grew 72%.
Lium operates as Subnet 51 on Bittensor, renting NVIDIA B300, H200 and RTX 5090 GPUs billed by the second.
Providers keep 95% of rental fees plus subnet emissions and idle-capacity bonuses.
Lium, a GPU rental service operating as Subnet 51 on Bittensor, recorded $2.70 million in customer GPU rentals over 31 days, with its latest-week spending pace annualizing to $34.6 million. The dashboard data, covering October 10 through November 9, showed 2,394 billed customers and 29,300 rentals. The figure reflects customer spending on the subnet, not collected network revenue, and excludes emissions and wash volume.
The distinction matters. The $34.6 million headline is a run rate, not a full year of booked revenue. It takes the most recent seven days of spending — roughly $95,000 per day — and multiplies it across 365 days. The average across the full 31-day window was about $87,000 per day, which annualizes closer to $31.8 million. Anyone quoting the larger number is describing a single week's momentum, not a year of contracted business.
What do the underlying numbers show?
The month-over-month growth is real, even if the annualized framing flatters it. Spending rose 87% from $1.45 million in the prior 31-day period. Billed customers increased 72%, rental volume grew 62%, and existing customers spent 46% more than before. That last figure is the strongest signal in the dataset: retention and deepening usage, not merely new sign-ups, drove the increase.
In plain terms, Lium added customers faster than it added rentals per customer early on, and then existing users expanded their spend — a pattern more consistent with genuine demand than one-off bursts.
How does the service actually operate?
Lium rents NVIDIA GPU capacity and bills customers by the second. Its listed high-utilization hardware includes NVIDIA B300, H200 and RTX 5090 GPUs. About 77.7% of rentals began through command-line tools, software development kits or application programming interfaces rather than the web app. That skew toward programmatic access suggests the customer base is dominated by developers and automated workloads, not casual browser-based users.
The provider economics are straightforward. Providers retain 95% of rental fees and separately receive subnet emissions and bonuses for idle capacity. Those emissions and idle-capacity payments are rewards distinct from rental income, meaning provider returns exceed what customer spending alone would support — a structural subsidy worth remembering when evaluating the subnet's sustainability without emissions.
How does this compare with network-wide estimates?
An earlier annualized revenue estimate for the entire Bittensor network put the figure at about $70 million across multiple subnets. Compute and infrastructure services accounted for roughly 78% to 82% of verified revenue in that estimate, with Lium named among the leading contributors alongside Targon and Engy. Other subnets offer services such as editable CAD files for 3D printing and high-volume token processing.
The scope of the two figures differs materially. The $70 million network estimate aggregates multiple subnets over a broader verification method; Lium's $34.6 million covers one service and annualizes a seven-day spending window. If Lium alone accounts for half of the network estimate, that implies compute dominates Bittensor's verified revenue to an even greater degree than the 78-82% range suggests — or that the two measurements are not directly comparable. The second reading is more likely.
The verification gap deserves attention. Neither figure has been audited, and the dashboard data is self-reported by the subnet. The exclusions — emissions and wash volume — are disclosed, but the extent of related-party rentals or Sybil-adjacent demand is not addressed in the data. A 46% increase in spend per existing customer is encouraging; it would be more convincing with independent confirmation that those customers are external.
The forward question is whether the latest-week pace holds. A run rate built on seven days can compress as quickly as it expanded, and the next dashboard update — roughly a month out — will show whether the $95,000 daily figure was a trend or a spike.
via tokenpost.com (Source)
More from Elena Vasquez
Show full bio
Staff writer covering regulation, policy and their impact on crypto and web3 industry at Mempool Brief.
1,075 articles