NVIDIA and Palantir are expanding their artificial-intelligence partnership into one of the most operationally important parts of the global economy: supply chains.
The companies announced Thursday that they have developed a new “sovereign AI” stack combining NVIDIA’s Nemotron open models with Palantir’s Foundry, Artificial Intelligence Platform and Ontology software. The system is being deployed first inside NVIDIA’s own supply chain before the companies attempt to extend the technology to customers across manufacturing, pharmaceuticals, retail, agriculture, energy, healthcare, automotive, aerospace, technology and government.
The commercial opportunity could be substantial if companies adopt the system broadly, but investors should separate that potential from what has actually been announced. NVIDIA and Palantir did not disclose a contract value, revenue forecast, number of committed customers or financial targets tied to the expanded partnership.
Instead, the announcement strengthens an increasingly close relationship between the world’s largest AI-chip company and one of the fastest-growing enterprise software businesses, with both attempting to move AI beyond chatbots and coding tools into complex decisions that directly affect manufacturing capacity, inventory and production.
NVIDIA Is Using the Technology on Its Own Supply Chain
The most notable part of the agreement is that NVIDIA is not simply selling the technology alongside Palantir.
It is becoming the first major deployment.
NVIDIA operates a sprawling global supply chain involving millions of individual parts, thousands of suppliers and manufacturing partners around the world. Producing one complete rack-scale AI system requires coordinated supplies of processors, memory, networking equipment, power systems, cooling equipment and mechanical components.
The scale becomes particularly clear with NVIDIA’s next-generation Vera Rubin systems. The company says each Vera Rubin rack involves approximately 1.3 million parts moving through its supply chain.
That creates an unusually complicated allocation problem.
A shortage of one seemingly small component can delay completion of a much more valuable AI system. Producing more GPUs does little good if NVIDIA cannot secure enough memory, networking equipment, cooling systems or power-related components to assemble finished products.
The new Palantir-NVIDIA system is designed to bring those pieces together inside a single operational environment.
Palantir Foundry organizes supply-chain data, while Palantir’s Ontology connects data to real-world objects, relationships and business processes. NVIDIA’s Nemotron models can then be customized using that proprietary operational information.
The companies say the resulting models can identify emerging constraints, recommend actions, explain trade-offs and help supply-chain teams evaluate different allocation decisions.
Human planners retain final authority.
That last point is particularly relevant for enterprise customers. The companies are not pitching an autonomous AI system that independently redirects billions of dollars of materials. They are attempting to create software that can process a much larger collection of operational information and present decision-makers with better options.
NVIDIA’s own engineers describe the objective as reducing the time between a semiconductor wafer leaving production and the finished system producing its first AI token.
AI Is Being Trained on Human Supply-Chain Decisions
NVIDIA provided more detail Thursday about how the internal system works.
The company built what it calls a Digital Supply Chain Intelligence command center using Palantir Foundry. NVIDIA’s cuOpt optimization software performs mathematical modeling to determine how materials should be distributed across manufacturing sites.
But NVIDIA found that its experienced human planners could sometimes outperform the mathematical optimization model.
The reason was information.
Experienced employees were incorporating details such as emails from suppliers, weather forecasts, geopolitical events and conversations with manufacturing partners — information that was not captured inside the traditional mathematical model.
NVIDIA and Palantir are now trying to turn that human expertise into something AI models can learn from.
NVIDIA says Nemotron 3.5 Lightning was post-trained using historical allocation decisions, the reasoning behind them and the eventual outcomes. Palantir’s software provides the governance layer around the process, while NVIDIA’s NeMo tools prepare data and help continuously improve the models.
The concept is financially interesting because companies frequently lose operational knowledge when experienced employees leave.
If AI systems can capture not only what planners decided but why they made those decisions — and then compare those choices with what eventually happened — companies may be able to preserve expertise that previously existed largely inside individual employees’ heads.
Whether the technology produces measurable productivity gains across other companies remains to be proven.
For NVIDIA, however, the internal deployment gives the partnership a significant real-world test.
Why Supply Chains Are a Natural AI Market
AI spending has so far been concentrated heavily around model training, cloud infrastructure and workplace applications.
Supply chains represent a different category.
They involve enormous datasets but also require constant decisions under changing conditions.
Manufacturers need to determine which factories receive scarce components. Retailers have to position inventory before demand appears. Pharmaceutical companies must coordinate sensitive production and distribution networks. Automakers depend on thousands of components arriving in the correct sequence.
Unexpected events can disrupt those plans quickly.
A factory shutdown, weather event, geopolitical conflict, shipping delay or supplier failure can force companies to reallocate inventory across an entire network.
Traditional supply-chain systems already perform optimization. The NVIDIA-Palantir approach attempts to combine those mathematical tools with AI models capable of incorporating less-structured information.
NVIDIA CEO Jensen Huang described supply chains as the “operating system of the physical economy,” arguing that AI infrastructure itself demonstrates how complicated these networks have become.
That complexity has become particularly visible during the AI infrastructure boom.
NVIDIA’s customers are spending hundreds of billions of dollars building data centers, yet every rack still depends on physical manufacturing capacity and components produced by companies across multiple countries.
Advanced GPUs require fabrication primarily from Taiwan Semiconductor Manufacturing. High-bandwidth memory comes from suppliers including SK Hynix, Samsung Electronics and Micron. Servers require networking equipment, power systems and cooling infrastructure from another group of vendors.
AI demand can therefore exceed NVIDIA’s ability to recognize revenue if bottlenecks elsewhere in the supply chain prevent complete systems from reaching customers.
Improving those logistics could have a more direct financial impact than many workplace AI applications because faster production can potentially translate into faster deliveries and revenue recognition.
The New System Is Built Around ‘Sovereign AI’
The second major part of the announcement concerns where enterprise data is stored and who controls it.
Palantir and NVIDIA are marketing the system around the concept of sovereign AI.
Companies can customize Nemotron models using their own internal information while retaining control over the models, data and deployment environment.
The stack can run on company-owned infrastructure rather than requiring businesses to transmit sensitive operational information to a third-party public AI service.
That can matter enormously for supply chains.
Procurement agreements, component shortages, production capacity, manufacturing yields and supplier negotiations can contain some of a company’s most sensitive competitive information.
A semiconductor manufacturer would have little incentive to place confidential production constraints inside an AI environment if it could not control where that information was stored or how it was used.
Palantir CEO Alex Karp has increasingly emphasized this issue, particularly for governments and highly regulated industries.
The new supply-chain system can run on premises using systems from Cisco and Dell Technologies, while customers can also deploy it in colocation facilities or cloud environments through providers including Rackspace and Nebius.
The underlying reference design is Palantir’s Sovereign AI Operating System architecture, supported by NVIDIA infrastructure.
That deployment flexibility could make the platform more attractive to defense contractors, government agencies, healthcare organizations and manufacturers that cannot simply upload sensitive operational data to a public cloud model.
The Partnership Has Been Building for Nearly a Year
Thursday’s agreement is an expansion of a relationship that started well before this week’s announcement.
NVIDIA and Palantir announced a broader operational-AI partnership in October 2025.
That agreement combined NVIDIA accelerated computing, CUDA-X libraries and Nemotron models with Palantir’s Ontology and AIP.
Lowe’s was among the early companies applying the combined technology to supply-chain logistics.
The partnership moved deeper into sensitive environments this June.
Palantir introduced an AI system for U.S. government agencies using NVIDIA Nemotron models that can run inside closed infrastructure. Agencies can train the models using their own information while retaining ownership of the customized models and their weights.
That arrangement was designed for government agencies and critical-infrastructure operators unable to place sensitive information inside conventional public AI services.
Thursday’s announcement takes the same basic architecture and expands it toward commercial supply chains.
The progression is strategically significant.
NVIDIA increasingly wants its AI models and software embedded inside the applications enterprises use, not merely inside the hardware running those applications.
Palantir wants its Ontology and AIP to become the operating layer through which companies deploy AI against proprietary data.
Each company therefore extends the other’s reach.
NVIDIA supplies the accelerated computing, optimization software and AI models.
Palantir supplies the enterprise data architecture, governance and operational workflow.
Palantir Has a Lot Riding on Enterprise AI Growth
The expanded NVIDIA relationship arrives while Palantir is reporting some of the fastest growth in the software industry.
Second-quarter revenue reached $1.94 billion, up 93% from a year earlier.
Operating income was $912 million, representing a 47% GAAP operating margin. Adjusted operating income reached $1.19 billion, or 62% of revenue, while adjusted free cash flow totaled $1.22 billion.
Palantir also raised its full-year outlook.
The company now expects 2026 revenue between $8.15 billion and $8.158 billion, with U.S. commercial revenue exceeding $3.424 billion. That would represent growth of at least 134% in the company’s U.S. commercial business.
Management expects adjusted free cash flow between $4.5 billion and $4.7 billion for the year.
Those numbers have changed how investors evaluate Palantir.
The company was once viewed primarily as a government and defense contractor with a growing commercial business.
AIP has increasingly pushed commercial customers toward the center of its growth story.
Supply-chain AI could broaden that opportunity because operational planning exists across nearly every major industry.
The question is whether the NVIDIA partnership generates incremental customers and contract value substantial enough to matter against Palantir’s rapidly expanding revenue base.
Neither company provided those figures Thursday.
NVIDIA Is Trying to Turn Its AI Lead Into a Full Software Ecosystem
For NVIDIA, Palantir represents another route for expanding beyond GPU sales.
The company’s latest financial results remain dominated by infrastructure.
Fiscal second-quarter revenue reached $96.2 billion, more than double the year-earlier level. Data Center revenue surged 117% to $89 billion.
NVIDIA reported a 75% gross margin, while adjusted diluted earnings reached $2.22 per share.
Those numbers demonstrate why supply-chain optimization matters internally.
At NVIDIA’s current revenue scale, production constraints can potentially represent billions of dollars of delayed business.
Demand for AI computing has continued increasing as hyperscalers, governments and AI laboratories build increasingly large infrastructure clusters.
NVIDIA is therefore attempting to improve the supply chain responsible for delivering its own products while simultaneously turning the software used to do so into something other companies can adopt.
That creates a potential feedback loop.
NVIDIA can test the technology against one of the world’s most demanding technology supply chains, refine it internally and then offer the architecture to outside companies alongside Palantir.
If successful, NVIDIA gains another enterprise software distribution channel for Nemotron, cuOpt and its broader AI stack.
Palantir gains one of the most visible demonstrations possible for its software.
NVIDIA’s Latest AI Strategy Is Increasingly About Open Models
The partnership also reinforces another shift in NVIDIA’s strategy.
The company has been aggressively expanding Nemotron as an open-model alternative for enterprises building AI agents.
Unlike closed frontier models where companies typically access the model through a provider’s service, Nemotron models can be customized and deployed inside infrastructure controlled by customers.
That does not mean NVIDIA is attempting to replace every closed AI model.
Instead, it gives businesses another option when they require control over model weights, data or deployment.
NVIDIA has recently integrated Nemotron with enterprise software companies across engineering, cybersecurity and government.
Palantir has become one of the more important partners in that effort.
In June, NVIDIA announced that Palantir, CrowdStrike and several enterprise software providers were building long-running AI agents using Nemotron and NVIDIA’s Agent Toolkit.
Supply chains could be particularly well suited to that strategy because companies often need highly specialized models rather than a general-purpose chatbot.
A model trained to understand NVIDIA’s component allocation decisions does not need to know everything about the internet.
It needs to understand NVIDIA’s suppliers, factories, constraints and production priorities exceptionally well.
That is the economic argument behind specialized enterprise AI.
The Market Did Not Treat the Announcement as an Immediate Earnings Event
Shares of both companies traded lower Thursday despite the partnership announcement.
NVIDIA was down roughly 2% during the session, while Palantir also fell around 1% to 2%.
The moves occurred during a broader technology selloff as Brent crude surged above $107 a barrel and the 10-year Treasury yield climbed toward 4.93%, increasing pressure on highly valued growth stocks. The S&P 500 and Nasdaq were also lower.
There is no clear evidence that investors were selling either stock because of the partnership.
The absence of financial terms also limits how much investors can immediately incorporate into earnings estimates.
A partnership can create strategic value without materially changing next-quarter revenue.
For Palantir, investors will eventually want evidence that NVIDIA’s deployment helps produce additional large commercial contracts.
For NVIDIA, the more direct benefit may initially come from improvements inside its own supply chain rather than software revenue generated from outside customers.
Dell and Cisco Could Gain From On-Premise Deployment
The agreement could also create opportunities for infrastructure companies involved in deploying sovereign AI.
Dell Technologies and Cisco support the Palantir Sovereign AI Operating System reference architecture.
Customers that choose to keep models and operational data inside their own facilities may need additional servers, networking equipment and accelerated computing infrastructure.
That could create hardware demand for vendors supporting the deployments.
Nebius and Rackspace are positioned on the hosted side of the architecture, providing alternatives for companies that want dedicated or cloud infrastructure without operating everything themselves.
Those companies should not be treated as guaranteed financial beneficiaries from Thursday’s announcement.
No purchase commitments or revenue amounts were disclosed.
But their involvement shows that Palantir and NVIDIA are attempting to build an ecosystem around sovereign enterprise AI rather than a product that only operates inside one vendor’s cloud.
What Investors Need to See Next
The partnership now has to move from technical capability to measurable economics.
The first benchmark will be NVIDIA itself.
If the company can demonstrate that Palantir and Nemotron materially reduce component shortages, accelerate production decisions or shorten the period between wafer production and completed AI systems, it would provide unusually strong validation for the product.
NVIDIA’s enormous scale makes even modest improvements potentially meaningful.
The second benchmark will be outside customers.
Palantir and NVIDIA say companies across agriculture, manufacturing, pharmaceuticals, retail, technology and government can deploy the same architecture, while additional applications are possible in energy, healthcare, automotive and aerospace.
Investors should watch future AIPCon presentations, contract announcements and Palantir earnings calls for evidence that customers are moving from demonstrations into paid deployments.
Contract values would provide the clearest signal.
Neither NVIDIA nor Palantir said Thursday how much customers will pay, what percentage of the economics each company receives or how quickly implementations can reach production.
Palantir’s commercial revenue growth makes that evidence particularly important. With U.S. commercial revenue already expected to rise at least 134% this year, new partnerships need to become substantial deployments to materially change the company’s growth trajectory.
For NVIDIA, the financial threshold is even higher.
A company generating more than $96 billion in quarterly revenue requires enormous new businesses to noticeably alter its earnings profile.
That means the near-term investment case is less about partnership revenue and more about what the deployment says about NVIDIA’s broader strategy.
The company that built the dominant hardware platform for training AI is increasingly trying to control more of the software stack around how businesses actually use those systems.
Palantir is trying to establish itself as the enterprise layer that connects those models to proprietary data and real-world operations.
Supply chains provide the two companies with an unusually demanding place to prove that combination works.
If NVIDIA can use the system to manage millions of components and 1.3 million parts inside each Vera Rubin rack, Palantir and NVIDIA will have a compelling reference customer when they take the technology to other industries.
The next step is proving that those customers are willing to pay for it at scale.