Founders club offering GB300 cluster
A Bay Area founders group is advertising an NVIDIA GB300 cluster — 32 GPUs and pricing aimed at 5K+ startups — pitching better price and real supply as an alternative to flaky cloud instances. That kind of grassroots cluster plumbing is already surfacing where developers are fed up with unreliable infra. (x.com)
A post from the Bay Area Founders Club account run by Dr. Paul Fang offered a GB300‑based 32‑GPU cluster aimed at early startups with a $5,000 entry price and an explicit pitch of “real supply” over cloud spot reliability. (x.com) NVIDIA describes the GB300 NVL72 as a liquid‑cooled, rack‑scale system that unifies 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs into a single NVLink domain for test‑time scaling and reasoning workloads. (nvidia.com) Microsoft announced and deployed an NDv6 GB300 VM series and a production GB300 NVL72 cluster with more than 4,600 GB300 GPUs in Azure—configurations Microsoft says are purpose‑built for OpenAI‑scale inference. (azure.microsoft.com) Specialist providers are already packaging GB300 access: Together.ai lists reserveable GB300 NVL72 clusters, Lambda published a Mountain View deployment of GB300 NVL72 systems (including a hydrogen‑powered facility), and Massed Compute offers custom multi‑node GPU cluster builds for training. (together.ai) OEM and enterprise routes remain the primary sales channel for GB300 hardware—HPE advertises an NVIDIA GB300 NVL72 integrated system and NVIDIA markets DGX GB300 as a rack‑scale enterprise product with vendor services. (buy.hpe.com) Industry price comparisons and startup programs show friction in on‑demand clouds: Microsoft for Startups documents quota and access delays for high‑end GPU VMs, while market guides note spot/discount instances can be interrupted or require quota approvals that take days, which vendors of fixed‑supply clusters cite when selling guaranteed capacity. (learn.microsoft.com)
