PantheonGet Early Access

How data-center GPUs are sold

TL;DR

Data-center GPUs reach buyers through four routes: OEM direct, two-tier distribution (a broadline distributor selling to a reseller who sells to you), a solution provider or systems integrator, and the secondary market. The structural point most buying guides miss is that the channel is organized by function, not by brand — a single broadline distributor carries NVIDIA alongside the major server OEMs rather than being tied to one brand, and the OEM partner programs layered on top govern margin, deal registration and allocation rather than acting as separate places to buy.

On this page

You do not buy brand by brand

Sourcing accelerated compute looks like a brand problem — apply to NVIDIA, then Dell, then HPE, then Supermicro. That is not how the market is arranged.

Enterprise hardware moves through two-tier distribution. The manufacturer sells to a small number of authorized distributors; the distributors sell to resellers; the reseller sells to the end buyer. NVIDIA’s own partner taxonomy makes the split explicit — a distributor is the partner type authorized to distribute NVIDIA products to resellers, while solution providers and integrators sit on the tier that faces the customer (NVIDIA Partner Network).

The consequence is the useful part. Because that tier is defined by function rather than by brand, broadline distributors are multi-vendor by construction: the same firm authorized to distribute NVIDIA products also carries the major server OEMs — Dell, HPE and Lenovo — along with AMD. One relationship at that tier reaches essentially the whole accelerated-computing catalog, which is why brand-by-brand effort is spent against a channel that was consolidated on purpose.

The four routes to a data-center GPU

Every path from a manufacturing line to a rack is one of four shapes. They differ less in what you receive than in who carries the engineering, the credit risk and the wait.

RouteWho you transact withBest forTime costPrice position
OEM directThe manufacturer’s own enterprise sales teamVery large, named, multi-rack deploymentsLongest — qualification first, then a production slotAnchored to list; movement comes from volume
Two-tier distributionA reseller, who sources from a broadline distributorStandard configurations, repeat buying, mixed-vendor buildsShortest for catalog parts; allocation-bound for scarce onesDistributor cost plus the reseller’s margin
Solution provider / integratorA firm that designs, builds and supports the clusterBuyers who need compute, power, cooling and support designed togetherLonger — a design phase precedes the orderHighest, because engineering and support sit inside the number
Secondary marketA dealer, or the prior owner directlyPrior-generation parts, urgent timelines, capacity that already existsShortest — the hardware exists todaySet by scarcity and condition, not by a price list

Partner programs govern margin, not access

Layered on top are the OEM partner programs — the NVIDIA Partner Network, Dell’s partner program, HPE Partner Ready, Lenovo 360, Supermicro’s channel program, and AMD’s partner hub. They are frequently mistaken for purchase channels. They are not.

A partner program is an accreditation layer, and there is no advertised membership fee at entry level. It governs the margin a partner earns, deal registration (claiming a named opportunity in order to protect the margin earned on it), co-marketing, and allocation priority on scarce parts. It does not replace the distributor account through which product is bought and invoiced.

The tiers are earned, not purchased. NVIDIA’s runs Registered → Preferred → Elite, with progression driven by certified staff, training and revenue attainment; every other major program follows the same shape under different names. For a buyer, a seller’s tier signals volume history and technical certification — not whether they can obtain the hardware.

What each route costs you in time

Price differences between routes are usually smaller than buyers expect. Time differences are not.

  • Allocation, not price, is the binding constraint on current-generation parts. When a flagship platform is supply-constrained, the question is which route holds allocation — decided by vendor relationship and registered demand, not willingness to pay.
  • Configuration drives lead time more than brand. A liquid-cooled rack-scale system carries a different build schedule from air-cooled nodes of the same generation, whoever sells it.
  • The secondary market trades price certainty for time certainty. The hardware exists, so delivery is logistics — but condition, runtime hours and warranty position must be established per unit.
  • Integration work is either priced in or it is yours. A solution provider’s number looks higher because it contains design, deployment and support that otherwise become your own project.

Which route fits which buyer

A buyer purchasing a known configuration, repeatedly, is best served by the two-tier path — fastest, most price-transparent, and most competitive, since several resellers can quote the same distributor stock. A buyer standing up a first cluster, or one whose site constraints are unresolved, is better off with a solution provider: the hard part is not the part number. A very large, well-defined requirement has the leverage to engage an OEM directly, and a buyer whose constraint is the calendar looks at the secondary market — trading provenance diligence for hardware that already exists.

Most real programs use more than one — current-generation capacity through distribution, prior-generation systems from the secondary market, on entirely different clocks.

To compare the systems themselves rather than the routes, see the data-center GPU catalog and the GPU comparison tool.

Frequently asked questions

What is the difference between an NVIDIA distributor and a reseller?

A distributor is authorized to supply resellers rather than end buyers, while the reseller is the customer-facing tier that quotes and invoices you — the full comparison, including where a dealer fits, is set out in distributor vs reseller vs dealer.

How do you buy NVIDIA data-center GPUs?

In practice, through a reseller or solution provider that sources from an authorized distributor. Direct purchase from NVIDIA is reserved for very large named deployments, and most enterprise volume moves through the two-tier channel instead. The route you pick should follow the requirement: a known configuration bought repeatedly suits a reseller quote, a first cluster with unresolved power and cooling suits a solution provider, and an urgent timeline often points at prior-generation systems on the secondary market.

Who sells H100 systems?

No single company. H100-class HGX systems are built by server OEMs and integrators — Supermicro, Dell, HPE and Lenovo among them — and reach buyers through authorized distribution and the resellers and solution providers those distributors supply. Because H100 is now a prior-generation part, a meaningful share of supply also moves through the secondary market as operators rotate to newer platforms.

Do you have to be an NVIDIA partner to buy data-center GPUs?

No — partner status matters to the firms selling the hardware, not to the organization deploying it, and what the badge actually governs is covered in how to become an NVIDIA partner.

Is buying direct from the OEM cheaper?

Usually not, and it is rarely the question that decides the outcome. Distribution margin is thin, and the OEM direct route carries qualification requirements and a production slot that the two-tier path can sometimes beat outright. On supply-constrained parts, the route that holds allocation delivers sooner regardless of which route quotes lower.

What is deal registration?

Deal registration is the mechanism by which a partner claims a specific named opportunity with the manufacturer. Once registered, that partner receives protected margin on the deal. It does not confer exclusivity — other partners can and do requote the same customer, but they do so from a weaker cost position and have to give up their own margin to compete. For a buyer it explains a common experience: quotes from several resellers converge closely, because only one of them is working from the protected buy price.

Related

New surplus vs refurbished vs used

New surplus is equipment that was built but never put into service — zero operating hours, factory condition, sold outside the original order because a project was cancelled, respecified, or over-ordered. Used equipment has run. Refurbished equipment has run and then been worked on. The gap that matters to a buyer is not cosmetic: it is warranty status, documented runtime hours, and who held title before you.

Read →

What is GPU allocation?

GPU allocation is the practice of distributing scarce accelerators against committed forward supply instead of selling them on demand. When a generation is spoken for before it is built, the buying question stops being "what does it cost" and becomes "who gets a slot". Priority follows volume history, partner tier, forecast commitments, and strategic importance — which is why a quoted lead time is mostly a position in a queue rather than a manufacturing time.

Read →

Distributor vs reseller vs dealer

In two-tier IT and electronics distribution, a distributor is a wholesaler: it buys in volume, warehouses stock, extends credit, and sells to the trade rather than to end customers. In industrial and power equipment the same word means something else — there a distributor is the manufacturer's regional sales and service organization, and it does sell end users directly. A reseller (or solution provider) sells to end customers and usually holds no stock at all; a dealer is the franchise-like middle case, with an assigned territory, a service obligation, and a required parts inventory. Nearly everyone searching "how to become a distributor" is describing the reseller role.

Read →

How gas turbines and gensets are sold: the buying channels

There is no self-serve door for a large gas turbine. A new machine is a direct capital sale from one of a handful of manufacturers, negotiated against a production slot booked years out; packagers and authorized service providers sit alongside that; the parts, repair and long-term-service aftermarket runs for decades after the sale; and existing units move through an intermediated used and surplus channel. Reciprocating gensets are the exception — series-built catalog products sold through authorized dealer and distributor networks. This page maps the channels from the buyer's side.

Read →

Buying vs Leasing vs Cloud GPUs

The hinge is sustained utilization. Owned hardware costs the same whether it runs at 20% or 95%, so ownership only pencils when demand is steady and high; cloud bills for what you consume, which is why it is genuinely the right answer for spiky, uncertain or short-lived workloads. Leasing sits in the middle, converting a large purchase into a predictable payment. Lead time, depreciation, residual value, control and data residency then decide the edge cases. Many organizations land on a hybrid: own the steady baseline, burst to cloud.

Read →

Share this page

Last updated