--- title: "Databricks charges $7 per H100 GPU-hour, 2.4 times the market rate, and caps a node at eight GPUs" description: "AI Runtime, Databricks' serverless GPU service, went generally available on October 21 at $7.00 per H100-hour, about 2.4 times the market rental rate. The docs cap a node at eight GPUs and warn regional capacity can run out." publisher: "The Inference" section: "Data Center" published: 2026-10-26T14:29:58.400Z modified: 2026-10-26T14:29:58.400Z canonical: https://theinference.org/article/databricks-charges-7-per-h100-gpu-hour-2-4-times-the-market-rate-and-caps-a-node-at-eight-gpus language: en keywords: "Databricks, GPUs, Pricing, Nvidia, Cloud Computing" --- # Databricks charges $7 per H100 GPU-hour, 2.4 times the market rate, and caps a node at eight GPUs > AI Runtime, Databricks' serverless GPU service, went generally available on October 21 at $7.00 per H100-hour, about 2.4 times the market rental rate. The docs cap a node at eight GPUs and warn regional capacity can run out. [Databricks](https://theinference.org/markets/companies/databricks), the private data-platform company [valued at $190 billion by its August investors](https://techcrunch.com/2026/08/13/databricks-wanted-to-raise-1b-investors-wanted-15b-it-settled-on-5b-at-a-190b-valuation/) [1], [moved AI Runtime to general availability on October 21, 2026](https://www.databricks.com/blog/ai-runtime-generally-available) [2]. The service rents serverless [NVIDIA](https://theinference.org/markets/companies/nvidia) GPUs for fine-tuning open models inside the lakehouse, the governed store where its enterprise customers already keep their data, with NVIDIA's NeMo containers and the new Nemotron 3.5 models available from day one [3]. The Inference [tracked Databricks' paper here on October 2](https://theinference.org/article/the-newest-disclosed-buyer-of-databricks-190-billion-paper-is-a-korean-vc); the new fact this week is the service's first public unit price, and the fine print under it [3]. ## The showcase ran 16 GPUs. The docs cap a node at eight The showcase customer made the pitch. Kostia Omelianchuk, an applied research scientist at the email app maker Superhuman, moved a training run from one node of eight A100 GPUs to 16 H100s; training time fell from about seven days to 48 hours, a 3.5-times speedup [2]. The quote does not separate the newer chip's gain from the second node's [4], and the second node is the product's open seam. Databricks' documentation, updated October 20, the day before launch, says all AI Runtime accelerators provision a single node, capped at eight GPUs [4]. ## Seven dollars an hour, on hardware Databricks does not own List price on [AWS](https://theinference.org/markets/companies/amazon), Premium plan, US East: $7.00 per H100 GPU-hour and $2.50 per A10, both still tagged Preview after general availability [4]. At list, Superhuman's run is 768 GPU-hours, about $5,376, with two nodes billing $112 an hour [4]. The open market rents the same chip for less: [Silicon Data's standardized index of specialized GPU clouds](https://www.silicondata.com/products/silicon-index/h100) read $2.89 per H100-hour on October 25 [5]. Databricks' list price is about 2.4 times what the open market charges for identical silicon, and 52 cents below the market rate for NVIDIA's newer B300, $7.52 [5]. *Chart: **Databricks' $7 H100-hour sits above the open market for the same chip** Bar chart comparing H100 per-GPU-hour prices: Databricks AI Runtime list at $7.00 against Silicon Data market indices from $2.89 for an H100 to $7.52 for a B300* *Databricks AI Runtime list price (AWS, Premium plan, US East, tagged Preview) against Silicon Data standardized market rental indices; US dollars per GPU-hour, October 25, 2026. [4][5]* ## The docs, not the blog, describe the product The launch post says workloads "scale to hundreds of GPUs reliably", backed by 100,000-plus GPU-hour runs at hundreds of customers including [Robinhood](https://theinference.org/markets/companies/robinhood), Merck and [FactSet](https://theinference.org/markets/companies/factset) [2]. The hardware table has no row beyond eight GPUs [4]. The docs warn capacity "may be constrained or unavailable in your region", say high-demand runs can land in other regions at data-transfer cost, and allow 14-day CLI jobs whose restart needs hardware again [4]. Databricks does not own the silicon: its chief executive said in August it carries multibillion-dollar cloud commitments with AWS, [Microsoft](https://theinference.org/markets/companies/microsoft) and Google [1]. Rental rates are already moving, with the H100 index up 6.2 percent in a week and the B300 up 10.6 percent [5]. If wholesale hardens, Databricks absorbs the spread, re-tags the price, or declines supply under [the footnote that displayed pricing "does not guarantee product availability"](https://www.ctol.digital/news/databricks-ai-runtime-ga-single-node-docs-multi-node-customer/) [4]. ## Tubi is happy. The analysts are split The counterweight is real usage. Tubi's principal machine-learning scientist is "very happy with the compute availability" [2]. [Analysts credit AIR's integration with Unity Catalog](https://www.techtarget.com/data-technologies/news/366652428/Databricks-latest-improves-AI-memory-infrastructure-for-agents), Databricks' permissions layer, as the actual differentiator, while flagging that available GPU capacity is the open question given limited supply, and that most enterprises may never need training compute at this scale [6]. ## Watch the Preview tag Two signals say whether the $7 anchor holds: whether the Preview tags come off the prices, and the B300 waitlist, the one inventory line in the docs [4]. A two-node job started on launch day that runs the full 14-day cap hits it on November 4, and the restart is another bet on capacity being there [4]. ## Takeaway Databricks' first public AI Runtime price, $7.00 per H100 GPU-hour, is about 2.4 times the open-market rate, and its docs cap a node at eight GPUs, the seam under the 16-GPU showcase run; whether the Preview tags come off tells you if the premium holds. ## Sources 1. [TechCrunch, "Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation.", 13 August 2026](https://techcrunch.com/2026/08/13/databricks-wanted-to-raise-1b-investors-wanted-15b-it-settled-on-5b-at-a-190b-valuation/) 2. [Databricks Blog, "AI Runtime is Generally Available", 21 October 2026](https://www.databricks.com/blog/ai-runtime-generally-available) 3. [Databricks Blog, "Databricks AI Runtime and NVIDIA make it easier to customize open models with enterprise data", 21 October 2026](https://www.databricks.com/blog/databricks-ai-runtime-nvidia-train-fine-tune-models) 4. [CTOL Digital Solutions, "Databricks AI Runtime caps a node at eight GPUs. Its showcase ran 16", 21 October 2026](https://www.ctol.digital/news/databricks-ai-runtime-ga-single-node-docs-multi-node-customer/) 5. [Silicon Data, "H100 Rental Price Index (SDH100RT)", reading of 25 October 2026](https://www.silicondata.com/products/silicon-index/h100) 6. [TechTarget, "Databricks' latest improves AI memory, infrastructure for agents", 22 October 2026](https://www.techtarget.com/data-technologies/news/366652428/Databricks-latest-improves-AI-memory-infrastructure-for-agents) --- Databricks charges $7 per H100 GPU-hour, 2.4 times the market rate, and caps a node at eight GPUs — The Inference. Canonical: https://theinference.org/article/databricks-charges-7-per-h100-gpu-hour-2-4-times-the-market-rate-and-caps-a-node-at-eight-gpus