Speak With An Expert

LIQID as a Service

Break through the AI memory wall with the industry’s only single-tenant, memory-scaled infrastructure.

Dedicated, single-tenant AI infrastructure with up to 30 GPUs and 160 TB of memory on one server. Fully managed by LIQID as a service.
LIQID as a Service rack with pooled GPUs and CXL memory connected to a single server
5
–
7
x
more tokens per second from the same GPUs*
30
GPUs
+
160
TB
of CXL memory on a single server
1
customer
per rack. Never shared.
Air cooled
about 24 kW per fully built node

*Based on initial testing using the same pooled-memory technology that powers Micron's Abaco reference design deployed at Pacific Northwest National Laboratory. Results vary by model and workload.

Diagram of AI workloads outgrowing the GPU and memory limits of a single server

AI outgrew the server.

A typical AI server tops out at eight GPUs and a few terabytes of memory. Past that point, teams split work across servers, spill data to storage that is about 100 times slower, or let GPUs recompute work they already finished. Each workaround costs speed, and many teams buy extra GPUs just to get more memory.

Buying more GPUs that ultimately sit idle is the most expensive way over the memory wall.

What you get

Rack-scale memory in one server, run for you.

More tokens from the GPUs you have.

LIQID’s fabric and LIQID Matrix software connect one server to pools of GPUs and CXL memory. Models, context, and data stay in memory, and initial testing delivered 5-7x more tokens per second from the same GPUs.

Private by design.

Every deployment is single tenant. You get a whole or half rack that is never shared, with the security boundary at your network edge. Bring your own models and run your own stack.

Fully managed.

LIQID handles setup, validation, hardware maintenance, and fault monitoring. No capital project or procurement cycle.

Conforms to today’s datacenter power and cooling envelopes.

The typical datacenter rack draws about 11 kW, and 87% of datacenters have no rack at 50 kW or more.** A fully built LIQID node draws about 22 to 24 kW and is air cooled, so it runs without liquid cooling or new construction.

Nodes start at 20 GPUs and 10 TB of CXL memory and scale to 30 GPUs and 160 TB. GPU options: NVIDIA H200 NVL, NVIDIA RTX PRO 6000 and AMD MI350P.

**Uptime Institute, 2026 Global Data Center Survey.

Use cases

Built for work that has outgrown one server.

Industries that can’t wait for the answer

Quantitative trading

Backtest against years of tick-level data held in memory, and move a signal into production while it still has an edge.

Fraud prevention

Keep the full fraud graph in memory and score every transaction against every connection.

Insurance underwriting

Run risk models on full portfolios instead of samples, and rerun scenarios as conditions change.

Pharma and life sciences

Search for targets across whole genomic and screening datasets, not slices.

Workloads that outgrow one server

Long-context inference and AI agents

Give the KV cache room to grow, so GPUs produce new tokens instead of redoing finished work.

Large models on one server

Run up to 30 GPUs on a single host, instead of splitting a model across a network.

Graph analytics and HPC simulation

Keep the whole graph or simulation in up to 160 TB of memory behind one host.

Talk to our team.

Tell us what you’re running and where it hits limits. A LIQID specialist will follow up within one business day to talk through sizing, pilot options and timing.

FAQ

Questions, answered

Is the hardware shared with other customers?

No. Every deployment is single tenant. You get a whole or half rack, and the security boundary sits at your network edge.

What do I manage, and what does LIQID manage?

LIQID handles setup, validation, hardware maintenance and fault monitoring. You run your models and your stack.

Which GPUs are available?

NVIDIA H200 NVL, NVIDIA RTX PRO 6000 and AMD MI350P.

Where does it run?

LIQID operates the infrastructure in existing datacenters, colocation sites, and neocloud, and service provider edge environments.

When can I start?

General availability is planned for Q1 2027.

Your models need more memory. Let’s talk.

See how much more your GPUs can do with rack-scale memory on one dedicated server, fully managed by LIQID.

Talk to sales