FlopSource

Daily News

AI and Compute: Daily Briefing (Nov 05, 2026)

By Tyler Babin · FlopSource Research

The H100 floor rebounds to $1.99 at Voltage Park, Google backstops Anthropic's $19bn TeraWulf lease, and Oracle plans multi-gigawatt Rubin superclusters. Plus: OpenAI pledges no-NDA data center deals after voter backlash, and Smartbird teams with Supermicro for managed single-tenant clusters.

Published Nov 05, 2026

1. H100 market floor rebounds to $1.99 at Voltage Park; Google holds at $5.38, Oracle at $10.00

As of November 5, the cheapest on-demand H100 SXM anywhere in FastGPU's live feed is $1.99 per GPU-hour at Voltage Park, up from the $1.79 floor recorded on November 4. Google Cloud is unchanged at $5.38 on demand and $1.09 on Spot, so Google's on-demand rate is 2.7x the floor. Oracle Cloud is also unchanged: $10.00 per GPU-hour on demand for an H100, with no preemptible H100 offered at all, which is 5.0x the $1.99 floor. The B200 floor is unchanged at $4.89 at DeepInfra, against Google's $1.74 Spot rate and Oracle's $14.00 on-demand rate. The H200 floor is $1.71 at Vast.ai, against Oracle's $10.00 on-demand rate. Operator read: yesterday's dip did not hold. The open-market floor is back where it sat on November 3, and the hyperscaler premium is unchanged - what moved was one marketplace listing, not the market.

Source: FastGPU, Google Cloud GPU Pricing in 2026 · Nov 05, 2026; FastGPU, Oracle Cloud GPU Pricing in 2026 · Nov 05, 2026

2. Google backstops Anthropic's 20-year, $19bn lease for 401 MW at TeraWulf's Kentucky campus

Google has agreed to backstop Anthropic's obligations under a 20-year, $19 billion lease covering 401 megawatts of computing capacity at TeraWulf's Kentucky campus, according to a TeraWulf filing disclosed on Wednesday and reported by CoinDesk on November 5. The arrangement is meant to help TeraWulf secure financing for the project, which is expected to begin delivering capacity in late 2027, with deliveries continuing through the second quarter of 2028. If Anthropic defaults on its rent or becomes insolvent, Google can step in as tenant and take over the lease, or, under certain conditions, pay a termination fee in exchange for a proportional ownership stake in Raylan Data, the TeraWulf subsidiary operating the project - not in TeraWulf itself. A termination payment equal to 30% of the project's total development cost would, for example, give Google a 30% stake. The agreements were signed on November 4, Google's support for each of the three leased sections starts only once the final data hall in that section is delivered, and the obligations end if Anthropic holds an investment-grade credit rating for at least three months. Operator read: credit support is now part of the compute stack. A miner-turned-developer can finance 401 MW because two larger names stand behind the rent, in sequence.

Source: CoinDesk, Google could take ownership stake in TeraWulf's AI project if Anthropic defaults · Nov 05, 2026 (TeraWulf filing Nov 04, 2026)

3. Oracle and Nvidia expand Vera Rubin on OCI to multi-gigawatt scale; Vera CPUs now available

Nvidia and Oracle announced plans to expand Oracle Cloud Infrastructure Superclusters with Nvidia Vera Rubin NVL72 systems to multi-gigawatt scale, in a press release published November 5. Vera Rubin NVL72 and Nvidia Vera CPUs are now available on OCI. The Superclusters combine NVL72 rack-scale systems - 72 GPUs per rack connected through NVLink - with Oracle Acceleron networking using Nvidia Spectrum-X Ethernet with Spectrum-6 switches and BlueField-4 DPUs, plus gigawatt-class power and cooling, scaling across hundreds of thousands of GPUs. OCI also plans to introduce Nvidia HGX Rubin NVL8, an eight-GPU Rubin server in an x86 configuration with two 64-core Intel Xeon 6 CPUs, 4 TB of DDR5 memory and 6.4 Tbps of RDMA networking. The release says OpenAI has trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 Nvidia Blackwell GPUs in Abilene. No Rubin pricing was disclosed. Operator read: Rubin is now an orderable OCI line, not a roadmap slide, but the multi-gigawatt figure is a plan. Judge it on delivered racks and published rates, neither of which is in this announcement.

Source: StorageNewsletter, reporting the Nvidia and Oracle press release · Nov 05, 2026

4. OpenAI launches National Community Compact: no NDAs, no resident electricity costs

OpenAI launched a National Community Compact on November 5, committing to transparent data center development after widespread voter opposition to AI infrastructure projects. The framework eliminates nondisclosure agreements in community negotiations and guarantees residents will not shoulder electricity costs for OpenAI facilities. OpenAI also pledges to publish data on noise, water consumption and energy use before breaking ground, reduce power use during peak demand periods, contribute electricity back to the grid when possible, set local hiring and workforce development goals, and publicly report progress. Chief global affairs officer Chris Lehane said the framework follows months of conversations with residents and officials in Georgia, Ohio, Texas and other states. The context is scale: at least 45 projects valued at $88 billion faced disruption from local opposition in the second quarter of 2026 alone, according to Data Center Watch. Operator read: social license is now a schedule risk on the same level as power interconnection. A developer that cannot show its community math up front should expect its timeline to slip.

Source: Emergent, OpenAI Pledges Data Center Reforms After Voter Backlash · Nov 05, 2026

5. Smartbird and Supermicro team up on fully managed, single-tenant AI clusters

Smartbird announced on November 5 a collaboration with Supermicro to deliver dedicated, fully managed AI infrastructure for enterprise workloads. Smartbird will design, deploy and operate single-tenant clusters built on Supermicro's computing platforms and rack-scale systems, including its Data Center Building Block Solutions, deployed on-premises or in colocation facilities for customers that need data isolation, security and operational control. Smartbird, launched in June 2026, sells the model as infrastructure tailored per customer - processor, capacity and deployment scale matched to the workload - without the customer building or operating the underlying stack. No pricing, cluster sizes or named customers were disclosed. Operator read: this is the managed end of the market formalizing again, the same pattern as Hetzner's fixed monthly bare metal yesterday. The offer is accountability and sovereignty, not a cheaper GPU-hour, and it will be priced that way.

Source: GlobeNewswire, Smartbird press release · Nov 05, 2026

Sources

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