Nvidia fell 1.4% as worries about funding AI spending weighed on the group. The decline does not establish a setback in Nvidia’s own demand, and holding up better than peers for one day has not repaired its recent pullback.
Stop guessing. Get the call, a three-month target price, the best and worst cases, and a source for every claim.
NVIDIA supplies cloud providers, AI developers and enterprises with the infrastructure to train and run AI models and process other demanding workloads. It combines graphics processing units (GPUs), networking equipment and other hardware with CUDA and related software, selling products ranging from modules to complete rack-scale systems. Customers buy directly or through server makers, system integrators and distributors. Data-center sales provide most of its revenue; NVIDIA also charges for software licenses such as NVIDIA AI Enterprise.
Outside data centers, NVIDIA sells GeForce GPUs for gaming PCs and NVIDIA RTX GPUs for professional workstations. These graphics businesses are much smaller than its data-center business.
Revenue is concentrated among a small group of direct buyers and end customers purchasing through partners. Sales depend on those customers having the funding, power and facilities to deploy the systems. NVIDIA outsources chip production and assembly, relying on suppliers including TSMC and Samsung for wafers and SK Hynix, Micron and Samsung for memory. U.S. export controls restrict which products it can sell and to whom; NVIDIA says it is effectively shut out of China's data-center computing market because it cannot deliver a competitive product approved by both governments.
The risk section is the longest, dullest part of the filing. This keeps only what can really hurt you: where it hits, how hard, when it plays out.
Every company claims an edge. Each one is checked against hard numbers like margins and returns: which hold, and which line gives way first.
Beyond this list, watch Huawei's competition in China's restricted AI computing market and Samsung's chip manufacturing and memory supply, both of which matter to NVIDIA's results.
AMD challenges NVIDIA for spending on AI accelerators, data-center processors and networking, and discrete gaming graphics cards, selling chips directly and through PC and server makers. Watch its ROCm software support for generative AI: it is part of AMD's effort to make its computing products an alternative to NVIDIA's.
NVIDIA depends on TSMC to turn its chip designs into semiconductor wafers and provide the advanced packaging essential to its AI chips. TSMC's updates on advanced manufacturing and packaging capacity help investors judge the production capacity behind NVIDIA's computing platforms.
Broadcom competes for the large cloud providers' infrastructure budgets that also fund NVIDIA's platforms: its custom AI accelerators and networking chips offer alternatives to NVIDIA's GPU computing and data-center networks.
NVIDIA buys memory from SK hynix, the market leader in high-bandwidth memory, or HBM, that supports high-performance AI accelerators. Watch SK hynix's progress in tailoring HBM to customers and passing their testing and approval requirements, which matter to usable memory supply.
NVIDIA relies on Micron for memory, including the high-bandwidth memory used in AI accelerators. Micron's cloud and data-center memory sales offer a read on server demand and the increasing memory needs of AI workloads that also use NVIDIA computing infrastructure.
Microsoft brings NVIDIA demand by buying computing infrastructure for Azure AI services, but also develops internal chips that can replace NVIDIA processors. Its cloud services therefore create both a customer relationship and a route to substitution.
Amazon supports NVIDIA sales by buying GPUs for AWS cloud services, while its own Trainium and Inferentia chips give customers alternatives for training AI models and running them to produce answers.
Alphabet creates demand for NVIDIA GPUs through Google Cloud and AI workloads, but its internally developed AI processors, called TPUs, compete for the same computing demand.
Oracle turns demand for its cloud AI services into purchases of NVIDIA GPU infrastructure, using that infrastructure to let customers train large AI models and run them to produce results.
Super Micro carries NVIDIA products into AI servers and liquid-cooled rack-scale systems by buying its processors and networking components. It sells those systems to data-center and cloud customers, linking their infrastructure purchases to NVIDIA demand.
Dell brings NVIDIA demand from two product lines: it buys GPUs for AI-optimized servers and graphics processors for PCs and workstations. Its sales connect NVIDIA to both computing-infrastructure buyers and commercial and consumer PC customers.
Arista competes with NVIDIA for the networks connecting AI data-center computers, selling Ethernet systems to cloud providers and enterprises against NVIDIA's Ethernet, InfiniBand and NVLink offerings. Watch its AI networking business for evidence of which network designs those buyers choose.
Marvell competes with parts of NVIDIA's computing and networking platform by selling customers custom compute chips and connections between data-center equipment. Watch demand for its customized products, which reflects buyers' appetite for infrastructure tailored to their own requirements.
Vertiv shares NVIDIA's exposure to power-dense AI data-center expansion: it sells the power and liquid-cooling equipment needed to operate NVIDIA's high-density systems. Its power-management and liquid-cooling business shows another part of the infrastructure required to put that computing capacity to work.
Cerebras offers NVIDIA's model-developer and enterprise customers an alternative: AI systems built around a wafer-sized processor, or cloud computing capacity for training and running models. Its customer deployments show where buyers are choosing a different computing architecture rather than NVIDIA-based infrastructure.
Intel competes with NVIDIA for data-center AI accelerators, processors and networking, as well as PC graphics, selling chips directly and through equipment makers and distributors. Its data-center AI accelerator and networking business is the part most directly relevant to NVIDIA's computing-infrastructure sales.
HPE brings NVIDIA GPUs to customers through its AI servers, but also sells networking products that compete with NVIDIA's networks. Its server relationship supports NVIDIA demand while the network side gives customers a competing choice.
NVIDIA has a multiyear agreement to buy advanced lasers and optical networking products from Coherent for AI infrastructure. Coherent's manufacturing expansion is worth watching for the supply of optical products that connect large AI computing systems.
Credo and NVIDIA serve the same AI infrastructure buildout: Credo sells high-speed cables and connectivity chips to large cloud providers and emerging AI clouds, while NVIDIA supplies computing platforms. Credo's cable and connectivity demand offers a view of the links needed as those AI clusters expand.
Nebius turns expansion of its AI cloud services into purchases of NVIDIA GPU infrastructure. It sells customers computing capacity for training and running AI models, making its cloud expansion a source of NVIDIA demand.
IREN buys NVIDIA GPU systems and networking to sell AI cloud capacity to large cloud providers, model developers and enterprises. Watch the mix of systems it installs and orders: its portfolio includes multiple NVIDIA generations alongside AMD MI350X, so not all of its compute expansion goes to NVIDIA.
Space Exploration Technologies operates AI infrastructure containing approximately 325,000 NVIDIA GPUs for internal workloads and cloud capacity sold to Anthropic. Watch the Anthropic capacity agreement: after its initial three-month period, either side can terminate on 90 days' notice, making contracted demand for that GPU capacity important to follow.
NVIDIA pays Fabrinet to manufacture and test products, making Fabrinet part of its outsourced production chain. NVIDIA accounted for 16.3% of Fabrinet's fiscal 2026 revenue; revenue from that customer is the line to watch for Fabrinet's NVIDIA-related manufacturing activity.
Lumentum shares in the high-speed connectivity spending needed as NVIDIA-based AI clusters expand, selling optical components for customers to integrate and complete networking systems. Its optical modules and optical circuit switches are concrete products to watch for that infrastructure buildout.
NVIDIA depends on CPU architecture licensed from Arm for its Grace processors and other system-on-chip products. Arm supplies the processor design technology NVIDIA uses, rather than manufacturing those chips.
Celestica shares NVIDIA's exposure to large cloud providers' AI infrastructure spending, designing and manufacturing networking and server equipment whose deployment expands alongside NVIDIA computing platforms. Its cloud-facing networking and server business offers a view of the surrounding infrastructure buildout, rather than a disclosed NVIDIA supply relationship.
Cisco competes with NVIDIA for data-center networking purchases through its AI networking chips and switches. Watch adoption of systems built on Cisco Silicon One, the chip architecture it uses across AI networking and switching products, for a concrete competing option to NVIDIA's networks.
Microsoft Corp earningsMSFT
Microsoft’s report will show whether cloud demand from Azure AI services continues to support its purchases of NVIDIA computing systems.
Read Azure cloud growth alongside infrastructure investment. Growth with continued investment would suggest Microsoft’s AI services are generating cloud demand that NVIDIA systems can serve; slower cloud growth, or greater use of Microsoft’s own chips, would mean less potential demand for NVIDIA’s systems. Microsoft’s overall results cannot establish how much NVIDIA equipment it bought, but they can show whether this customer-demand path is expanding or narrowing.
SK hynix Inc. earningsSKHY
SK hynix’s report will show whether high-bandwidth memory supply and customer qualification are keeping pace with demand for NVIDIA’s AI systems.
Read high-bandwidth memory shipments or sales alongside progress in customer testing and qualification. More supply that passes qualification would make usable memory more available for NVIDIA accelerators; if production rises but approval stalls, or deliveries are delayed, nominal capacity will not become usable components and system deliveries may remain constrained.
Vertiv Holdings Co earningsVRT
Vertiv’s report will show whether data-center customers are buying the power and liquid-cooling equipment needed to run NVIDIA’s high-density systems.
Watch demand in Vertiv’s power-management and liquid-cooling businesses. Stronger orders or sales would indicate customers are equipping data centers for denser computing, making NVIDIA systems easier to deploy; weaker demand could mean power or cooling infrastructure is slowing expansion and customers may delay installing NVIDIA systems.
Fabrinet earningsFN
Fabrinet’s report will show whether its business manufacturing and testing products for NVIDIA is growing or shrinking.
Watch Fabrinet’s reported NVIDIA-related revenue against changes in its overall manufacturing business. Rising related revenue would show more NVIDIA product activity flowing through this manufacturer; a decline would indicate weaker activity along this production link. Revenue also depends on production and delivery timing, so one quarter alone cannot establish whether end demand rose or fell.
Advanced Micro Devices earningsAMD
AMD’s report will show whether its AI accelerators and ROCm software are winning spending that might otherwise go to NVIDIA’s computing platform.
Watch AMD’s data-center AI business and whether ROCm support for generative AI is translating into adoption. If accelerator sales and software adoption both strengthen, customers are more willing to put some AI computing on a platform other than NVIDIA’s; limited adoption would mean less competitive pressure on NVIDIA’s computing platform from AMD.
Arista Networks earningsANET
Arista Networks’ report will show whether cloud providers building AI clusters are adopting Ethernet systems that compete with NVIDIA’s networking products.
Watch Arista Networks’ AI networking business and cloud providers’ adoption of its Ethernet systems. Growth and greater adoption would show AI clusters are still expanding, while buyers direct some network spending to designs that compete with NVIDIA’s; weaker business could signal slower cluster expansion, reducing one source of support for deploying NVIDIA computing systems.
Nvidia Corp earningsThis company
NVIDIA’s revenue is driven mainly by its Data Center business, and this report will show whether sales growth is still translating into profit.
Read Data Center revenue alongside gross margin, the share of sales left after product costs. Rising revenue with a stable or improving margin would show NVIDIA selling more computing and networking systems without costs eroding the added sales. Slower revenue with a falling margin, especially alongside inventory or purchase-obligation charges, would point to weaker demand or transition and supply-commitment costs squeezing the business’s returns.
