Nvidia H100 Rent Hits $2.88 an Hour, Up 44% Since December

Renting an Nvidia H100 from a neo-cloud provider cost $2.88 per GPU-hour on Silicon Data’s H100 rental index, the index page showed on 24 October 2026. On 9 December 2025 the same index read $2.00, Silicon Data wrote in a January blog post.

By our arithmetic that is a rise of 44% in about ten and a half months. Social Capital, the investment firm, put the rise at 45% in an X post on 24 October, using a reading of $2.90 from 23 October.

Business Pill · GPUS

A short explainer of why AI runs on GPUs. It teaches the general idea only and says nothing about rental prices.

What the Index Shows

Silicon Data publishes the H100 Neo-Cloud index daily as ticker SDH100RT. Its page says the index is a standardized, like-for-like hourly rate expressed in USD per GPU-hour, drawn from observations across neo-cloud providers and GPU marketplaces, with hyperscaler pricing published separately.

The published readings trace the climb: $2.00 on 9 December 2025 and $2.20 on 6 January 2026, from the January post; $2.47 on 8 April and $2.76 on 30 May, from the June post; $2.57 on 4 May and $2.75 on 27 July, from the August post; and $2.88 on the index page on 24 October.

The January post called the first step “a startling 10% price jump for H100 rentals between December 9, 2025 and January 6, 2026.” The June post described an “11.7% trough-to-peak swing” between 8 April and 30 May.

What Was Posted on X

Social Capital, the investment firm, posted that renting an H100 costs 45% more than in December, though it went on sale in 2022 and two newer generations have come out since, and argued that demand for compute is so far ahead of supply that older GPUs are getting more expensive.

Elon Musk, replying to Yuchen Jin (@Yuchenj_UW) on X, writes: “I think we’re going to see a lot of stranded compute in partially complete data centers.”

Yuchen Jin (@Yuchenj_UW), co-founder of Hyperbolic, relayed on 24 Oct a remark from an unnamed friend at a neolab: the lab locked in GPU capacity early and could resell it, as H100, B200 and B300 rental prices have gone up 2-3x. The quote is second-hand and unverified; compare it with the rental data in this piece.

Newer Chips Rent for More, and the Old One Still Rose

The same Silicon Data index list, read on 24 October, put the H100 next to newer chips and the hyperscaler tier:

Source: Silicon Data index list, read 24 October 2026.

Silicon Data’s August post said the H200 Neo-Cloud rate opened May at $2.70 and stood at $3.09 on 27 July, a 14.4% rise over twelve weeks. The H100 rate rose too, it wrote, “but half as much and all of it early.”

Social Capital’s post read: “It went on sale in 2022, and two newer generations have come out since. Yet, demand for compute is so far ahead of supply that older GPUs are getting more expensive.”

Connecting the Coverage

Our renewal-trap piece of 18 October reported an argument made on SemiAnalysis Ep. 037: B300s are sold out or six months out, and one-year H100 terms are back near first-half-2023 pricing, “like three bucks an hour.”

Our sellout piece reported that Amazon chief executive Andy Jassy said of AWS computing capacity, as Reuters reported on 23 October: “2027 is largely sold out.” Our Nvidia guarantee piece reported that Nvidia’s Form 10-Q lists guarantees capped at $105 billion for an OpenAI data centre campus in Ohio. And our power-plant piece reported AEP saying Amazon would cover the costs of a planned 960-megawatt plant if it goes ahead.

What is new here: a dated price series for the H100 itself, from one index provider, showing the rent on a 2022 chip rising through 2026.

The Structural Read

As we read it, the rise was not a single jump: the published readings show $2.20 in early January, $2.47 in April, $2.76 at the end of May and $2.75 in late July before $2.88 in October.

As we read it, the newer chips set the context: on the same index list the H200 reads $3.48 and the B200 $6.45 per GPU-hour, so the H100 is still the cheaper option even after its rise.

As we read it, the neo-cloud and hyperscaler tiers sit far apart for the same chip: $2.88 against $7.17 for the H100 on the 24 October index list.

Three Implications

FOR RENTERS The index read $2.88 per GPU-hour on 24 October, up from $2.00 on 9 December 2025; 44% is our arithmetic.

FOR HOLDERS OF OLDER CHIPS Our 18 October piece reported an argument that one-year H100 terms are back near first-half-2023 pricing.

FOR BUYERS COMPARING TIERS The 24 October index list shows the H100 at $7.17 per GPU-hour at hyperscalers against $2.88 at neo-clouds.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Supercycle.

The framework puts it this way: “Layer 5 — Compute Capacity. Hyperscaler GPU clusters, neoclouds, custom-silicon estates.”

As we read it, the H100 index is a price reading from that layer, from the neo-cloud part of it. The framework does not discuss this index.

What Is Not Established

The index covers the neo-cloud segment, not every contract; hyperscaler rates are a separate reading. The readings above are the ones Silicon Data published, so they do not show every move between them. The 44% figure is our arithmetic from two published readings; Social Capital’s 45% uses a 23 October reading we did not see on the index page. The index page also carries an undated FAQ answer with a lower H100 price; we used the reading dated 24 October at the top of the page. We did not contact Silicon Data or Social Capital.

The Bottom Line

Silicon Data’s H100 neo-cloud index read $2.88 per GPU-hour on 24 October 2026, up from $2.00 on 9 December 2025, a 44% rise by our arithmetic, while newer chips on the same index rent for more.