Blackwell Ultra
B300
Largest-memory option for frontier training
- Memory / GPU
- 288 GB HBM3e
- Node
- 8 × GPU
- Fit
- Train · Reason
US capacity · from January 2027
Reserved GPU capacity, placed at the right moment.
Oberth Compute matches AI teams with reserved NVIDIA H100, H200, B200 and B300 capacity in the United States, on terms from one month to five years.
Data center operators we work with
Capacity sourced across leading colocation and data center platforms.
01Capacity
Four generations.
One point of contact.
From proven Hopper systems to Blackwell Ultra, delivered as 8-GPU nodes. Tell us the workload and the date; we bring back the options.
Blackwell
B200
High-throughput training and inference
- Memory / GPU
- 180 GB HBM3e
- Node
- 8 × GPU
- Fit
- Train · Serve
Hopper
H200
More memory on the Hopper platform
- Memory / GPU
- 141 GB HBM3e
- Node
- 8 × GPU
- Fit
- Serve · Fine-tune
Hopper
H100
The proven workhorse
- Memory / GPU
- 80 GB HBM3
- Node
- 8 × GPU
- Fit
- Train · Serve
Long-term
3 – 5 years
Reserved capacity for teams planning their compute roadmap.
Short-term
1 month – 1 year
Flexible terms for training runs, launches and bursts.
Availability
From Jan 2027
Reserved US capacity. Need it sooner? Ask, and we will check.
02Capacity matcher
Shape your request
in thirty seconds.
Set the GPU, size, start and term. Send it, and we come back with matched options and pricing.
03Process
Three steps
to a signed term.
The Oberth effect: the burn is most efficient at the closest pass. We do the searching, so your team only engages at the moment that matters.
01
Share what you need
GPU type, node count, start date, term and any network or storage requirements.
02
We match the capacity
We line up options that fit, with location, ready date, interconnect and pricing.
03
Choose and sign
Compare the options, pick the one that fits and lock in your reserved term.
For AI teams
Compute on your roadmap, not on a waitlist.
Labs, startups and enterprises that need committed capacity for training or inference.
- Reserved nodes for multi-year plans
- Short terms for runs and launches
- One contact from request to signature
For capacity owners
Fill upcoming capacity with committed buyers.
Data centers, neoclouds and operators with blocks coming online.
- Long-term demand for new deployments
- Mutual NDA before specifics are shared
- Referral and co-brokering arrangements
04What a request looks like
Ask once.
Get options.
An illustrative exchange. Real replies include location, ready date, interconnect, terms and pricing for each option.
AI
An AI teamto Oberth Compute
We need 16 nodes of B200 for 12 months, starting Q1 2027. US preferred, InfiniBand required.
Sent with the capacity matcher
Oberth Computeto the AI team
Two options that fit:
OPTION A
16 × HGX B200 · InfiniBandReady Q1 2027
12-month term · US[pricing in reply]
OPTION B
16 × HGX B300 · InfiniBandReady Jan 2027
12-month term · US[pricing in reply]
Happy to set up calls with both providers this week.
Illustrative example
05Why a broker
Skip the search.
Most capacity is never listed publicly. A broker sees more of the market than any single sales team.
On your own
With Oberth
Finding supply
Contact providers one by one
One request, several options
Comparing
Different formats and quotes
Side-by-side terms
Hardware range
Whatever one provider has
H100, H200, B200 and B300
Terms
Fixed by the provider
1 month to 5 years
Your time
Weeks of calls
One conversation
06FAQ
Questions,
answered.
Which GPUs can I lease?
NVIDIA H100, H200, B200 and B300, delivered as 8-GPU nodes.
What contract lengths are available?
Long-term reservations of 3 to 5 years, and short terms from 1 month to 1 year.
When is capacity available?
Reserved US capacity is available from January 2027. If you need an earlier date, include it in your request and we will check what is possible.
Where is the capacity located?
In the United States.
Can teams outside the US lease capacity?
Yes, where US export rules allow it. Tell us where your team is based when you send a request.
Will you sign an NDA?
Yes. We are happy to sign a mutual NDA before either side shares specifics.
How do I get started?
Use the capacity matcher above, or email gliderhover1@gmail.com with your GPU type, size, start date and term.
07Contact
Let's place
your compute.
Tell us what you are building and when. We reply with options and pricing.
gliderhover1@gmail.com