OEM Server Manufacturers in 2026: Dell, Supermicro, Lenovo, HPE
Summary
- Dell, Supermicro, Lenovo and HPE are the largest branded OEM server manufacturers by revenue in IDC's 2Q26 tracker (13.4%, 6.1%, 5.1% and 3.5% share).
- ODMs such as Foxconn and Quanta build most servers by revenue, mainly for hyperscalers buying direct, so a neocloud usually buys from a branded OEM.
- Compare OEMs on cooling, warranty, rack integration and vendor financing for the same board. Your supplier and purchase order are core financing-review documents.
Introduction
The leading OEM server manufacturers in 2026 are Dell Technologies, Supermicro, Lenovo and HPE, ranked by 2Q26 server revenue in IDC's worldwide tracker. Each sells GPU servers built on NVIDIA's HGX B300 board, directly or through authorized partners.
An operator with a signed offtake contract still has to pick who builds the servers. That choice sets the price, the delivery date and the warranty. It also decides what goes on the purchase order a financing review examines.
IDC's Kuba Stolarski says "Demand is broadening beyond the largest hyperscalers toward specialized cloud providers (or neoclouds) scaling quickly." IDC also reports that GPU server average selling prices rose 43.6% year on year in 2Q26, from about $118,600 to about $170,200. It attributes the rise to DRAM and NAND pricing and component allocation.
The sections below explain who builds GPU servers, rank the OEMs on IDC data and compare them on the points a GPU buyer checks. They finish with how your supplier shows up in a financing review. Specifying the server itself is a separate topic, covered in how GPU servers are built.
Who Builds GPU Servers: OEMs, ODMs and NVIDIA's Own Systems
For a GPU cloud operator, the realistic suppliers are branded OEMs and their authorized partners. ODM-direct buying made up 53.9% of server revenue in IDC's 2Q26 tracker, and it is mostly a hyperscaler channel.
Four kinds of company sit between NVIDIA's board and your data center:
- OEM (original equipment manufacturer). Designs, brands, builds and supports its own server line under its name: Dell PowerEdge, Supermicro, Lenovo ThinkSystem, HPE ProLiant and Compute.
- ODM (original design manufacturer). Builds servers to a customer's design, often at hyperscale. Foxconn, Quanta, Wiwynn and Inventec are examples.
- Reseller. Sells another company's servers and adds no manufacturing, though it may add services.
- System integrator. Assembles, racks, cables and tests systems from several vendors before delivery.
The ODM channel is the largest by revenue, and its share is falling. IDC puts ODM Direct revenue at $89.7 billion in 2Q26, with its share down to 53.9% from 60.6% in 2Q25. Foxconn attributes its AI rack growth to "continued capital expenditure expansion among cloud service providers (CSP)" in its Q2 2026 results of Aug 12, 2026.
Those cloud service providers are the hyperscalers, which buy direct at a scale few operators match. For a neocloud, the practical route runs through a branded OEM or one of its partners.
NVIDIA's HGX boards go into servers that system builders design, while DGX is NVIDIA's own system configuration, as HGX vs DGX explained sets out. NVIDIA's certified systems list includes certified systems from Dell, HPE, Lenovo, Supermicro, GIGABYTE, Cisco, Fujitsu and others.
The Largest OEM Server Manufacturers in 2026
Dell led branded vendors with $22.2 billion of server revenue in IDC's 2Q26 tracker, about twice Supermicro's $10.2 billion. Dell, Supermicro and Lenovo roughly doubled revenue or more year on year, while IEIT Systems, in fifth place, shrank 8.1%.
| Vendor | 2Q26 revenue | Share | YoY growth |
|---|---|---|---|
| Dell Technologies | $22.2B | 13.4% | +165.4% |
| Supermicro | $10.2B | 6.1% | +97.2% |
| Lenovo | $8.4B | 5.1% | +99.6% |
| HPE | $5.9B | 3.5% | +46.0% |
| IEIT Systems | $4.0B | 2.4% | −8.1% |
| ODM Direct | $89.7B | 53.9% | Not shown |
Source: IDC, 2Q26 (April–June 2026). Shares are of total server revenue.
The company figures below come from fiscal quarters that end on different dates: Jul 31, 2026 for Dell and HPE, Jun 30 for Lenovo and Sep 30 for Supermicro. Don't compare them as if simultaneous. Orders, backlog and pipeline are company-reported measures of future demand, and none of them is revenue.
Dell Technologies
Dell reported $16.4 billion of AI server revenue in the quarter ended Jul 31, 2026. It booked $60.9 billion of AI orders and ended the quarter with a $95 billion AI backlog. The orders and backlog are not yet revenue. Dell also raised its fiscal 2027 AI server revenue guidance to $74 billion from $60 billion.
Dell supplies neoclouds directly. IREN said on Nov 3, 2025 that it agreed to buy GPUs and ancillary equipment from Dell for about $5.8 billion. In a Nov 25, 2025 announcement, Dell said IREN's Prince George site will be Canada's first cluster of Dell PowerEdge servers with GB300 NVL72.
The NVL72 is bought as a full rack of 72 GPUs, as what an NVL72 rack is explains. In a Jun 1, 2026 blog post, Dell says it was first to ship Vera Rubin systems, delivering them to CoreWeave. That is Dell's own claim.
Supermicro
Supermicro reported net sales of $19.9 billion for the quarter ended Sep 30, 2026, against $5.0 billion a year earlier. The results are unaudited and preliminary. Gross margin was 12.7%, and as of Nov 3, 2026 the company guided fiscal 2027 net sales of $72 billion to $82 billion.
Supermicro's Lambda case study says "Lambda selected Supermicro for its GPU-optimized servers." The systems it lists are HGX B200 and H200 servers, and the page carries no date.
Lenovo
Lenovo's Infrastructure Solutions Group revenue rose 98% to $8.5 billion in the quarter ended Jun 30, 2026, according to Lenovo's results. Lenovo also reported a $54 billion AI server pipeline. A pipeline counts potential deals before they become orders or revenue.
Lenovo says it rose to #2 globally in x86 server revenue, which is a company claim. IDC ranks Lenovo third among all server vendors in 2Q26.
HPE
HPE's earnings presentation shows Cloud & AI revenue up 25.4% to $9.0 billion in the quarter ended Jul 31, 2026. Server revenue rose 35.3% to $6.8 billion. HPE booked $2.4 billion in new AI systems orders, and its AI systems backlog rose to $6.8 billion. HPE notes that orders and backlog remain subject to adjustment, and neither is revenue.
Supply is the constraint HPE names. HPE CEO Antonio Neri said on the Q3 FY26 call in September 2026: "Supply will continue to be constrained, which means we're going to continue to run into high backlogs."
What Actually Differs Between OEMs for a GPU Buyer
On the same NVIDIA HGX B300 board, the four OEMs differ in cooling options, base warranty and how much of the rack they build and test before shipping. Supermicro's base warranty on GPU systems covers parts for one year, while HPE's XD685 covers parts for three.
Compare like with like: the same board, the same cooling and the same scope of delivery. Every model below carries the eight-GPU HGX B300 board, an SXM design, and SXM vs PCIe form factors covers the difference.
| OEM | Air-cooled B300 model | Liquid-cooled B300 model | Base warranty | Rack-scale integration |
|---|---|---|---|---|
| Dell | PowerEdge XE9780 (10U) | PowerEdge XE9785L (liquid-cooled, direct-to-chip) | Not stated on product pages | Dell PowerRack: tested as a complete system before the rack leaves the factory |
| Supermicro | SYS-822GS-NB3RT (8U; onsite service required) | SYS-422GS-NB3RT-LCC (4U; complete rack and onsite service required) | GPU systems: 3 years labor, 1 year parts, 1 year advance cross-ship | DCBBS: rack (L11) and cluster (L12) validation before shipping |
| HPE | Compute XD690 (10U) | ProLiant Compute XD685 (direct liquid cooling configurations) | XD685: 3 years parts, labor and onsite, next business day | HPE Services: installation and deployment for GB300 NVL72 |
| Lenovo | ThinkSystem SR680a V4 (8U) | None found | 3 years CRU and onsite, 9x5 next business day; these levels don't apply to GPU-related service requests | EveryScale: pre-integrated and tested at rack level in manufacturing |
Warranty on GPU parts is where the terms diverge most. Supermicro's warranty terms give GPU systems 3 years of labor, 1 year of parts and 1 year of advance cross-ship. HPE's XD685 QuickSpecs list "3-Year Parts, 3-Year Labor, 3-Year Onsite support with next business day response."
Lenovo's SR680a V4 product guide lists a three-year customer-replaceable unit (CRU) and onsite warranty at 9x5 next business day. The same guide says those service levels don't apply to GPU-related service requests on this server. Dell's product pages for these models don't state a base term in years, so get the GPU warranty term on every quote in writing.
Confirm whether a model requires a complete rack. Supermicro's SYS-422GS-NB3RT-LCC page says a complete rack and onsite service are required, so you can't order it as a standalone server. That changes the size of the purchase order and the delivery scope.
Rack-scale integration also differs. Dell PowerRack tests every component as a complete system before the rack leaves the factory. Supermicro's DCBBS adds rack (L11) and cluster (L12) validation before shipping.
Lenovo EveryScale hardware is pre-integrated and tested at rack level in manufacturing. HPE backs its GB300 NVL72 with installation and deployment through HPE Services.
Get delivery timing in writing as well. HPE's CEO expects supply to stay constrained, and your delivery date sets when the servers start earning under your offtake contract.
Vendor Financing from OEMs
The largest OEMs run their own financing arms, and Dell's tripled its new originations year on year in the quarter ended Jul 31, 2026. Dell's latest 10-Q puts those originations at $7.5 billion, against $2.4 billion a year earlier.
Dell Financial Services funded $11.9 billion of originations in fiscal 2026, according to Dell's annual report. It held a $14.3 billion portfolio of financing receivables at Jan 30, 2026.
HPE's 10-Q describes HPE Financial Services as offering leasing, financing, IT consumption, utility programs and asset management. The unit earned $0.9 billion of revenue in the quarter ended Jul 31, 2026.
Lenovo sells TruScale, which combines hardware, services and "flexible financing options." Lenovo Global Financial Services offers finance and lease, rental and pay-as-you-service plans.
In our view, vendor financing ties the loan to one supplier. Asset-backed financing underwritten on the offtake contract is a separate route, and it pays whichever OEM is on the purchase order. That matters when you split an order across OEMs or switch suppliers for an earlier delivery slot.
How Your OEM and Purchase Order Show up in a Financing Review
The purchase order is a core document in an asset-backed GPU financing review. It names the supplier, the exact model, the quantity and the price. Together with the supplier, it fixes what the loan buys and what secures it.
Large GPU purchase orders need credible proof that committed capital is available before an OEM builds and ships. Loan funds are often held in escrow for that reason. Escrow releases the capital once the release conditions are met: servers installed, burned in and independently verified.
Deployment cost is the eligible hardware on the purchase order: servers, networking (switches, optics and cabling), owned racks and power distribution units (PDUs). Soft costs are excluded, including installation, labor, shipping, taxes, software subscriptions and data-center build-out.
A financing review asks for these documents:
- Supplier and purchase order. The OEM or authorized partner and the priced order.
- System configuration and quantity. The exact server model and how many units.
- Colocation agreement. A signed agreement in a Tier III or Tier IV data center.
- Power and cooling readiness. Evidence the site can power and cool the servers on order.
- Offtake contract. The contract behind your revenue, or documented on-demand revenue.
- Equity contribution. Your share of the deployment cost.
Configurations vary by system builder, so the review is done on the exact model on the purchase order. A Dell, Supermicro, HPE or Lenovo B300 server each goes into the file as its own configuration.
GPU financing through USD.AI covers up to 80% of deployment cost, with the operator contributing at least 20% as equity. Loans carry fixed or floating rates over 3–5 year terms with straight-line monthly amortization, and financing starts from about $5 million. A three-month debt-service reserve is funded at closing.
The operator's equity is prepaid to the OEM with the order. The loan funds sit in escrow until the servers are installed, burned in and independently verified.
Illustrative example: our pricing research puts a complete eight-GPU B300 server at about $600,000 at Q4 2026 pricing, server-only and pre-tax. Sixteen servers come to about $9.6 million. At 80% of cost, that supports up to about $7.7 million of financing against at least about $1.9 million of equity, on server cost alone.
Networking, owned racks, PDUs and cabling on the same purchase order add to deployment cost. The operator pays installation and tax.
Conclusion
Dell, Supermicro, Lenovo and HPE are the OEM server manufacturers a neocloud can buy from. They held 13.4%, 6.1%, 5.1% and 3.5% of 2Q26 server revenue in IDC's tracker. ODMs mostly serve hyperscalers buying direct. Between OEMs, compare cooling options for the same board, GPU warranty, rack requirements and integration, and the financing route.
GPU financing through USD.AI works with whichever OEM is on your purchase order. USD.AI had reviewed over 700 GPU financing deals as of September 2026. Compute IQ scores each deal on the operator, the customer's credit and the offtake contract.
Bring your purchase order and offtake contract, and USD.AI will review the deployment with you.
FAQ
Who are the largest OEM server manufacturers?
Dell, Supermicro, Lenovo, HPE and IEIT Systems are the largest OEM server manufacturers in IDC's 2Q26 tracker, at 13.4%, 6.1%, 5.1%, 3.5% and 2.4% share. ODMs selling direct, mainly to hyperscalers, took 53.9%.
What is the difference between an OEM server manufacturer and a reseller?
An OEM server manufacturer designs, builds, brands and warrants the server, while a reseller sells it without manufacturing it. A reseller may add services on top of the OEM's product.
How much does a GPU server cost from an OEM?
A complete eight-GPU B300 server costs about $600,000 at Q4 2026 pricing, based on USD.AI's pricing research (server-only, pre-tax). IDC's average GPU server selling price was about $170,200 in 2Q26, across all GPU server types.
