PACINFRAX · RESOURCES
AI infrastructure you can own, place and run.
A roadmap connecting model APIs, dedicated compute and customer-owned GPU hosting. Choose the ownership model that fits your workload—not just another endpoint.
Planned services—not an available inventory, reservation or quote. External GPU integration is deferred to the owner-provisioned hardware phase. No hardware shipment, purchase or service activation is enabled here.
Rehearse partner requirements locally—categorical planning only, with no hardware submission or capacity reservation.
RENT COMPUTE
AI Cloud
Planned serverless and dedicated inference, GPU clusters, training and evaluation workflows.
- Versioned models, license checks and measured serving profiles.
- Explicit shared, dedicated, reserved and interruptible terms.
- Request-level usage and cost visibility with commercial billing kept separate from preview data.
KEEP OWNERSHIP
AI Colocation
Planned rack, private pod and managed hosting for customer-owned GPU servers.
- Hardware acceptance against verified power, cooling and network envelopes.
- Defined custody, physical access, remote hands and maintenance responsibilities.
- Separate agreements for space, energy, connectivity and optional managed operations.
SUPPLY CAPACITY
Partner Capacity
A planned route for approved asset owners to offer GPU capacity through PacinfraX.
- Qualified inventory and fenced allocation before a listing can become available.
- Separate supplier, customer and operator permissions.
- Explicit merchant, fees, usage evidence, disputes and settlement terms—not guaranteed utilization or returns.
Make the operating context part of the product.
These proposed capabilities define our design direction. They are not activated services or proven competitive superiority.
- Capacity Passport: a proposed record connecting GPU configuration, ownership, site, acceptance status and price version. Proposed supply never appears as available capacity.
- Cost Passport: compare compute, power, storage, network and service fees with the same units and assumptions. Unknown charges stay unresolved—not silently zero.
- GreenWindow: proposed opt-in scheduling of flexible jobs against measured renewable availability. Interactive workloads keep their agreed latency boundaries; energy savings remain unverified.
- Deployment choice: a consistent planned workflow across rented and customer-owned compute, with supported migrations and their limitations stated explicitly.
Vietnam infrastructure: evidence before promises.
The owner reports multiple self-designed data centers in Vietnam using solar and wind resources. Site commissioning, available capacity and energy measurements have not been independently verified for this preview.
Approved site evidence will distinguish renewable generation, grid imports and backup energy. No renewable percentage, certification, availability guarantee or automatic data-residency assurance is claimed.
A lower-price target, not a blanket promise.
We target 5–10% below selected like-for-like published offers, subject to measured cost and margin validation. PacinfraX sell prices are not published.
Comparisons must match model revision, hardware, precision, service mode, region, term and included fees. Expired promotions and quote-only offers are not valid automatic benchmarks. Open-weight models require release-specific license review; proprietary API access is a separate contractual product.
Prepare requirements locally—nothing is submitted automatically