Leasing Nvidia’s top-tier AI hardware simply got more expensive for the 17th day in a row. Since October 15, 2026, on-demand rental costs for Blackwell Ultra (B300) GPUs have actually pressed to an all-time high.
The figures originate from Compute Desk, which runs an index developed on confirmed rental deals from United States neocloud suppliers. Matthew Sigel, head of digital possessions research study at VanEck, flagged the pattern and indicated 2 motorists: strong need for open-source designs and supply restraints at Supermicro, which produces B300 systems.
What the numbers reveal
On-demand B300 leasings presently run around $7 to $10 per GPU-hour. Back in November 2025, the exact same calculate expense roughly $4.55 per hour.
That exercises to an increase of around 64%. Anybody who allocated AI training runs last winter season is most likely reviewing that spreadsheet.
The hardware itself has actually gotten better too. Mid-October reports show the B300 has actually gotten almost 40% in worth over the previous 3 months.
Calculate Desk’s index matters here due to the fact that GPU rental prices has actually traditionally been nontransparent. Lots of offers occur independently, and published sale price do not constantly match what clients in fact pay. An index anchored to validated deals offers the marketplace something better to a genuine rate signal.
The neocloud companies it tracks are a more recent class of cloud business. Unlike general-purpose cloud giants, they focus directly on leasing GPU capability for AI work, that makes their prices a relatively direct continue reading need for calculate.
Why the B300 commands a premium
The B300 is not simply a refresh of the previous Blackwell B200. It loads 288 GB of HBM3e memory and 8 TB/s of memory bandwidth.
The tradeoff is power. The B300 can prepare to 1,400 W, a significant action up that impacts how information centers prepare cooling and electrical capability.
Since Supermicro develops the B300 systems, any traffic jam on its end streams directly into rental accessibility.
Rental markets have actually revealed level of sensitivity to both Supermicro server schedule and waves of AI design releases. When a brand-new batch of designs lands and everybody wishes to run them simultaneously, a constrained supply pipeline has no place to provide other than rate.
Sigel’s focus on open-source designs deserves keeping in mind. Open designs can be downloaded and run by anybody with access to hardware, which spreads out need throughout a much larger swimming pool of purchasers than a handful of exclusive laboratories.
A detach in the general public markets
In spite of record leasing rates, public business connected to calculate leasing and mining are seeing blended market responses. Some are trading near their 52-week lows.
The leasing boom likewise does not map nicely onto any particular digital possession. Financiers trying to find a tidy crypto proxy for AI calculate need will not discover one in this information.
What this suggests
For business that currently own B300 capability, the present environment looks beneficial. Greater per hour rates and an almost 40% gain in hardware worth over 3 months reinforce both their earnings and their balance sheets.
For AI designers and start-ups leasing calculate, a roughly 64% dive because November 2025 raises the expense of training and serving designs, which might press some groups towards smaller sized designs, older chips, or longer-term agreements to secure prices.
Supermicro’s production speed is the clearest variable to keep track of. If its supply restrictions ease, rental costs might discover a ceiling. If they continue while brand-new design implementations keep showing up, the Compute Desk index has space to keep printing records.
Disclosure: This post was modified by Diego Almada Lopez. For more details on how we develop and evaluate material, see our Editorial Policy.
Discover more from PMN S.P.O.R.T.S - A PRIME MEDIA NETWORK BRAND
Subscribe to get the latest posts sent to your email.
