The AI-native marketplace for secondary NVIDIA GPUs.

Buy, sell, or auction nodes, with agents handling the sourcing.

NVIDIA DGX B300DGX B200GB300 NVL72GB200 NVL72HGX B200DGX H200HGX H200GH200 Grace HopperH100 SXML40SInfiniBand NDRSpectrum-XDirect-to-chip liquidRack integrationNVIDIA DGX B300DGX B200GB300 NVL72GB200 NVL72HGX B200DGX H200HGX H200GH200 Grace HopperH100 SXML40SInfiniBand NDRSpectrum-XDirect-to-chip liquidRack integration

From refresh plan to signed contract.

A quarter of sourcing work, done in days. Click any step to explore it.

You, in plain language

→ What the agents extract

✓

Systems

24× HGX B200 (8-GPU)

✓

Site

Austin, Texas

✓

Timeline

6 weeks

✓

Budget

~$9M

List inventory from the tools you already use.

Connect Claude, Cursor, or any MCP-capable agent, and your inventory lists itself from a spec sheet.

Read the setup guide

In your agent

You: List 12 of the attached DGX B200 systems at
     $265,000 each, 4 weeks lead time, Austin TX.

Agent: Created "NVIDIA DGX B200"
       in your inventory: 12 units, $265,000/unit,
       lead time 28 days, 3 photos attached.
       It is live in the catalog and matchable now.
  • Claude Desktop · Cursor · Codex · any MCP client
  • One URL and your supplier key
  • List · reprice · archive · attach photos

For neoclouds mid-refresh.

Data center operators and enterprise AI teams get the same agents, counterparties, and contract.

Moving a hall to the next architecture

Move a Hopper hall to Blackwell without a dark quarter.

  • Rack-scale GB200 / GB300 NVL72 and HGX B200 nodes from verified allocation
  • Delivery windows sequenced against the decommission, so racks never sit empty
  • Split across suppliers automatically when one can't cover the quantity

Selling down a generation

Sell down a generation while resale values hold.

  • Consign H100 and A100 fleets to the auction floor or list them in the catalog
  • Verified bidders only, a private reserve you control, and we bid for them by proxy
  • Your identity stays confidential until contract; we contract and fulfil

Adding proven capacity

Add proven capacity while the new allocation lands.

  • Tested and graded previous-generation systems, shipping from stock
  • Condition, generation, and lead time stated on every listing
  • Every order competed on landed price, lead time, and real availability

Standardizing the fleet

Standardize the fleet on the configuration you benchmarked.

  • Matched on GPU model, interconnect, memory, and chassis, never a rough equivalent
  • Lock lead times before an OEM allocation slips a quarter
  • Export-control, country-of-origin, and certification constraints applied up front

What you move to, and what you move out.

New systems from verified allocation, and tested used fleets with condition stated up front.

New · the next architecture

Sealed in crate, full warranty.

  • Current-generation DGX, HGX, and NVL72 systems from verified allocation
  • OEM warranty and support terms carried through to your contract
  • Lead times confirmed with the supplier before you sign

Used & refurbished · the last generation

Tested, graded, and the fleets changing hands.

  • Systems coming off other operators' floors as they refresh, tested before listing
  • Generation and condition stated on every listing, no surprises
  • Proven capacity below new pricing, and the lots you'll most often see at auction

From the catalog

On the marketplace now.

Live SKUs backed by verified supplier inventory. Indicative price bands shown; exact pricing is confirmed and competed during sourcing.

View the full catalog

Days

from first spec to a signature-ready contract, buying or selling

Verified

every fleet, every buyer, every lot: real systems, real lead times

Landed price

competed across sellers, disclosed before you sign

Stop sitting on last-generation silicon.

One request covers what goes out and what comes in. The agents take it to a contract you can sign.