AI and Compute: Daily Briefing (Nov 02, 2026)
Blackwell spot offers fell hard, Rubin reached the cloud shelf without a published price, and the day's two biggest capacity stories were about money and politics, not chips: a $2.6bn Saudi sovereign deal targeting 3 GW, and OpenAI preparing to rename its data centers as AI compute campuses.
1. B300 rental offers drop to $4.60 per hour, from $7.90 last month
GPU providers are now offering NVIDIA B300 access at $4.60 per GPU-hour with no usage commitment, down from $7.90 per hour last month, according to RightNow founder Jaber, whose comparison was posted on X on November 2 and reported by ChainCatcher. Treat this as a market signal, not a list price: it is a single founder's report of offers in the market, not a published provider rate card. The direction still matters for buyers. Spot and no-commitment Blackwell pricing is compressing fast even while reserved and hyperscaler rates stay firm.
2. Vera Rubin is on the cloud shelf, but nobody has published a Rubin price
NVIDIA's Vera Rubin platform has moved from announcement to early delivery: production shipments began in August, CoreWeave announced availability on September 30, Oracle Cloud announced general availability on October 27, and Google Cloud's A5X instances are due in November, per an eCorpIT FinOps analysis updated November 2. NVIDIA claims up to 5x the inference performance and up to 10x lower cost per token than Blackwell, with a 288 GB HBM4 Rubin GPU. None of the launch announcements carried Rubin instance prices, and AWS raised EC2 Capacity Blocks for ML pricing again on October 7 after earlier rises in 2026. Operator read: do not budget a Rubin discount until a provider publishes a rate you can rent at.
3. Aramco closes 49% stake in HUMAIN for about $2.6bn; 3 GW of AI capacity targeted by 2030
Saudi Arabia's Public Investment Fund and Aramco have closed the transaction that makes Aramco a 49% owner of HUMAIN, the kingdom's national AI company, with PIF keeping 51%. Aramco contributes about SAR 9.9 billion (about $2.6 billion) in cash in 2026, plus its own AI, compute and digital assets, under the joint announcement dated November 1. HUMAIN, launched in May 2025, works across data centers, cloud infrastructure and AI models, and plans to deploy 3 GW of AI data center capacity by 2030. The announcement does not state an implied valuation for HUMAIN or name the chip suppliers and customers for the planned capacity.
4. OpenAI will call its data centers AI compute campuses in a post-midterms policy push
OpenAI plans to start calling the data centers it builds AI compute campuses, as part of a policy and public relations push launching right after the November 3 midterm elections, Axios reported on November 1. The package pairs the new name with community compacts for host towns and proposed state legislation giving residents a say in how projects are built and what benefits they receive, according to chief global affairs officer Chris Lehane. OpenAI points to its Pike County, Ohio project as the model: a site with room for up to 8 GW of capacity, leased for 20 years from SoftBank's SB Energy, with Nvidia guaranteeing part of the lease. The operator read is simple: the name changes nothing about power draw or local electricity prices. The compacts and the legislation are the parts that decide who bears the cost.
5. Envision claims first commercial 800 V DC power system for a large AI data center
Envision Energy says it has deployed the world's first commercial integrated 800 V DC power system in a large-scale data center, at its Galaxy AI campus in Ulanqab, China. The system combines solid-state transformers and battery storage, converting 10 kV medium-voltage power to an 800 V DC bus, with millisecond-level coordination to absorb GPU load swings, according to the manufacturer. The Galaxy campus began operations in August 2026, has a planned capacity of 3 GW, and more than 80% of its power is expected to come direct from renewables. Envision plans to scale the approach through its Mission Gobi program, targeting 5 GW of green AI data center capacity in desert regions worldwide by 2030. Higher-voltage DC distribution is the same direction NVIDIA's rack roadmap is pushing: more compute per rack, fewer conversion losses.
Compare providers for your workload
Live pricing, hardware specs, and availability across the GPU clouds, neoclouds, and bare-metal providers in our directory.
Open the Directory