Jun. 17 at 10:02 PM
Baseten CEO: "If you go out right now saying you want a thousand GPUs, truly.. people are talking about Q2 of next year (12 months out, maybe 15 months out). We have a cluster.. in one of these clouds.. of B200s Blackwell chips. Our unit price right now is
$2.63/ hour.. that's up for renewal in Oct (2026) - they came to us already in May & said
$5.10 is the new price for next year...double"
--- Who is the CEO talking about???
Candidate 1:
$GOOGL GCP
- Baseten has a massive, highly-publicized partnership w/ GCP specifically for testing & optimizing
$NVDA Blackwell clusters on Google's A4 Virtual Machines. Google is one of the very few tier-1 cloud providers that has had physical Blackwell clusters live long enough for Baseten to already have an established contract up for renewal in Oct
Candidate 2: NeoCloud (
$CRWV or Lambda Labs)
-
$2.63/hour for a Blackwell B200 is an incredibly cheap "insider" or early-reservation price likely only available at a Neocloud
Why???
- Global median price is
$6.10 - at
$5.10 Baseten is still paying less
Note: Major players like Google Cloud (
$16.11) &
$AMZN AWS (
$14-
$14.20) charge for on-demand Blackwell chips. This is b/c they bundle elite interconnect systems (NVLink 5.0), vast system RAM, & massive vCPU power w/ the instance
- Lead times for procuring 1,000 GPUs have now typically stretched into the second quarter of next year, with wait periods ranging from 12 to 15 months. Under the dual pressure of rising prices and persistent supply constraints, the cost of AI inference is facing a substantial increase
Note: global confirmed stock availability for B200 listings sits at just 52%. This means nearly half of the advertised listings online are restricted or "provisioning-dependent
Note: As of Jan 2026, Basetan has completed its D+ round of financing & secured strategic investment from NVvidia