NVIDIA and OpenAI Announce ‘the Biggest AI Infrastructure Project in History’

I’m writing this as the creator of the announcement video and as someone who has been living and breathing the infrastructure needs of modern AI for years. In a live conversation with CNBC’s Jon Fortt, I sat down with Sam Altman and Greg Brockman of OpenAI at NVIDIA headquarters in Santa Clara to share a strategic partnership that I believe will materially change the trajectory of AI deployment worldwide. Together we revealed plans for OpenAI to deploy at least 10 gigawatts of NVIDIA-based systems — including our Vera Rubin platform — and for NVIDIA to invest up to $100 billion in OpenAI progressively as each gigawatt of capacity is deployed.

This article unpacks what we announced, why it matters, the technical and economic scale involved, the partners and players that make the project possible, how it complements cloud and edge strategies, and what I believe will come next. I’ll quote the key moments from our discussion with Sam and Greg, and add context to help readers — whether you’re a technologist, investor, policymaker, or curious citizen — understand the significance and implications of this unprecedented buildout.

📰 The Big Picture: What We Announced and Why It Matters

At a high level, the announcement is straightforward but enormous in scale: OpenAI will deploy at least 10 gigawatts of NVIDIA systems to build next-generation AI infrastructure, and NVIDIA intends to invest up to $100 billion in OpenAI progressively tied to the deployment of that capacity. I said plainly during the conversation: “This is the biggest AI infrastructure project in history. This is the largest computing project in history.”

Why is this necessary? In short, the world’s demand for frontier AI computing is exploding. ChatGPT and subsequent models have demonstrated capability and utility at a speed and scale that was unimaginable just a few years ago. As Sam said during our discussion, “Buildings and infrastructure are critical to everything we want to do. Without doing this, we cannot deliver the services people want.”

I couldn't have put it better: AI is ready to move from lab experiments into every industry, every application, and virtually every person’s daily computing experience. To realize that transformation reliably and safely, we must build infrastructure at a scale that matches the opportunity and demand.

⚙️ How Much Compute Are We Talking About?

Numbers like “10 gigawatts” and “$100 billion” are big and they can be easy to gloss over. To make this tangible, I walked through what 10 gigawatts roughly means and why it’s so consequential.

10 gigawatts of AI infrastructure equates to roughly 4–5 million GPUs, depending on the generation and type of accelerators deployed.
For context, that is approximately the amount of compute we shipped in one full year — in a single project, within a relatively short deployment window.
Scaling from hundreds of thousands to millions of GPUs is not merely a linear expansion; it is a monumental logistics and engineering effort spanning chip supply, system design, power distribution, heat management, data center construction, networking, and software orchestration.

In the interview I pointed out that even this massive number is still orders of magnitude short of what a future where every person has a persistent AI agent might ultimately require. I sketched that vision when I said: “You really want every person to be able to have their own dedicated GPU. So you're talking order of 10 billion GPUs we're going to need.” That underlines the point: this 10 GW project is foundational and historic, but it is also a first phase in a far larger, multi-year evolution.

💸 NVIDIA’s Investment: The $100 Billion Commitment

We announced that NVIDIA intends to invest up to $100 billion in OpenAI, deployed progressively as each gigawatt of capacity is completed. This investment is structured to align capital deployment with infrastructure rollout. The $100 billion number is intentionally large because the economics of building AI infrastructure at global scale require deep commitment from both technology and capital partners.

Why did we make such a commitment? Several reasons:

Speed: To enable OpenAI to secure systems and ramp capacity quickly.
Scale: To underwrite an infrastructure project larger than any prior computing endeavor.
Alignment: To align NVIDIA’s engineering and supply capabilities closely with OpenAI’s roadmap so we can iterate on platforms and systems together.

As I told CNBC, “$100 billion is a small dent in it” — meaning the overall capital and supply needs for the future of AI are vast, but this commitment is a meaningful, catalytic start.

🤝 Who’s Involved: Partners and the Ecosystem

This project does not exist in isolation. OpenAI has existing and growing partnerships with major cloud providers and infrastructure teams — Microsoft Azure, Oracle Cloud Infrastructure (OCI), CoreWeave, SoftBank, Stargate and others. What we announced with NVIDIA is additive to those relationships, not a replacement.

As Greg reminded viewers: “We are working together with Oracle to do a lot of the infrastructure builds with SoftBank and Stargate... But there's really been no partner like Jensen, like NVIDIA.” Sam echoed that sentiment, calling NVIDIA and Microsoft two of OpenAI’s most critical partners from the start.

One vital aspect here is specialization. Different partners play different roles:

Cloud providers (Azure, OCI, CoreWeave) deliver on-demand capacity and global reach.
Infrastructure builders and operators (Stargate, SoftBank) provide physical sites, operations, and project management expertise.
NVIDIA supplies the accelerated computing platform — chips, systems, and the Vera Rubin platform — tightly integrated with software stacks to maximize model performance.

Each partner contributes a piece of the puzzle, and the project scales by coordinating these roles across many regions and providers.

🔍 The Technical Core: NVIDIA Platforms and the Vera Rubin System

A headline from the description and our talks was that OpenAI will deploy NVIDIA systems including the Vera Rubin platform. Vera Rubin is our next-generation, hyperscale infrastructure architecture tuned specifically for massive AI workloads.

Key attributes of this platform include:

High-density GPU clusters optimized for training and inference at scale.
Advanced cooling and power delivery mechanisms to support dense deployments efficiently.
Optimized networking to minimize latency and maximize throughput for model parallelism and data movement.
Software integrations that align with the frameworks AI researchers and engineers use daily, including CUDA, cuDNN, and AI-specific orchestration software.

The combination of hardware and software integration is what enables orders-of-magnitude improvements in effective compute. As Sam said, we’ve been working together since 2016 — the tight alignment between model development and platform performance has been critical to OpenAI’s rapid progress.

🌐 Global Implications: Where Will This Capacity Live?

We discussed the global footprint and geopolitical considerations candidly in the interview. I emphasized that the United States should lead across the AI stack — chips, infrastructure, models, and applications. Building an American-centered tech stack offers strategic advantages and aligns with policy priorities. But this project will not be limited to the U.S.; AI infrastructure will expand globally.

My key points on geography and competitiveness:

We want the world to be built on the American tech stack, because we believe American companies should lead in chips and systems, models and applications.
That said, we will see AI infrastructure built in Europe, Southeast Asia, southern regions around the world — dispersed deployments are needed for latency, local regulation compliance, and market presence.
Restrictions and export policies (for example with certain technologies in specific countries) will shape how and where systems are deployed.

Ultimately, the goal is to diffuse capability quickly and responsibly so that industries across the world can adopt and benefit from AI innovations.

🧠 From Chat to Thinking Models: The Evolution of AI

During our discussion, Sam highlighted a critical gap between public perception of AI and what cutting-edge models are already capable of. Many people still equate AI with conversational chat or simple productivity tasks. But the frontier of AI is rapidly moving toward models that can reason, discover, and even assist in scientific breakthroughs.

Sam Altman: “AI is now outperforming humans at the most difficult intellectual competitions we have. For the first time with GPT-5, you're starting to see scientists saying AI is like making novel discoveries, small ones, but real ones.”

That’s a profound statement. The progression from generative AI (creating text and images) to reasoning models and now to “thinking” models means the compute needs and the kinds of workloads we support are also changing. The implication is twofold:

Models will require far more specialized, high-throughput inference and training capabilities.
Applications will multiply — from drug discovery to personalized education to enterprise automation.

When large models are unleashed with added inference capacity, they can do far more than passively answer questions; they can proactively act, discover, and optimize. That’s why infrastructure matters so much.

🏗️ The Logistics of a Global Infrastructure Build

I was candid about the sheer logistical complexity of a build like this. This is not simply ordering lots of chips and flipping a switch. It touches every part of the supply chain and requires coordination across multiple domains:

Chip fabrication and assembly — ensuring steady supply of accelerators at scale.
System integration — designing and validating dense server architectures.
Data center design — power planning, cooling systems, and mechanical engineering.
Networking — building low-latency, high-bandwidth interconnects for model parallelism.
Software and orchestration — deployment automation, cluster management, monitoring, and security.
Operations and talent — hiring and training engineers, technicians, and site staff to operate at hyperscale.

As I put it: “The amount of work it takes to build that out — the size and scale of these multi-square-mile gigantic things and the complexity at every level of supply chain.” It’s a systems problem at planetary scale.

📈 Demand Dynamics: Why People Worried (and Why They Shouldn’t Panic)

We touched on a market moment that many readers might remember — the so-called “deep seek moment” where the market reacted sharply to questions about compute growth and profitability. That panic was understandable, but it overestimated the idea that compute scaling would stop or that demand would vanish.

Sam explained that the world misjudged both the pace of AI model capability and the breadth of demand. He said: “People really want to predict the end of the compute scaling somehow. And then it turned out that people need a lot of AI and they need a lot of the frontier AI.”

Two dynamics to keep in mind:

Cost per unit of intelligence is falling. That means more access and broader reach for AI services.
At the same time, the frontier capability of top models keeps rising, creating new, high-value workloads that are compute-intensive.

Those two forces together drive massive incremental demand. This project is our response: to ensure capacity keeps pace with accelerating needs while continuing to lower cost and broaden access.

📡 Cloud and Edge: Complementary Strategies

We discussed the role of edge devices and national platforms in parallel with the hyperscale data center build. Apple’s recent iPhone announcements and the broader momentum toward more capable local AI on devices were directly addressed in our conversation.

My view is straightforward: we need both cloud and edge. I articulated that during the discussion: “For the kind of chat with ChatGPT about most stuff, you will be able to do a lot of that on your phone.” Sam added that OpenAI has released an open-source model (GPT-OSS) that can run on a laptop or phone, and the capability on devices is farther along than many expect.

But there are workloads that require gargantuan reasoning — drug discovery, scientific simulation, and other frontier research applications — that will remain in hyperscale data centers. So the future will be heterogeneous:

Edge-first for latency-sensitive, privacy-focused, or lower compute tasks.
Cloud/hyperscale for high-performance training, frontier inference, and aggregation of massive datasets.

In the interview I said, “You should expect that you'll see this whole ecosystem of models from us and from everyone else that is able to take advantage of what's out there. It's a little bit like when the App Store first came out.” That analogy captures why both new device platforms and enormous data centers will unlock new applications in parallel.

👨‍💻 Talent, Immigration, and the Human Factor

Building and operating this infrastructure depends on people. We talked about immigration policy and how it intersects with competitiveness. My perspective is clear: immigration is foundational to American innovation. I said, “We want all the brightest minds to come to the United States. Immigration is the foundation of the American dream.”

There’s a direct business case here as well: to design, build, and operate at this scale you need top-tier engineers, data scientists, and operators. Policies that streamline the ability to attract and retain talent will accelerate deployment and innovation.

Operational talent needs extend across many domains:

Hardware engineers for system design and deployment.
Data center build teams — civil, electrical, mechanical engineers.
Software engineers for orchestration, monitoring, and model-serving systems.
Data scientists and researchers to exploit the compute capacity for new science and products.

We will be working hard to attract that talent and to advocate for policies that ensure the U.S. remains competitive.

🔮 The Near-Term Horizon: What to Expect Next

During the talk I signaled that the 10 gigawatts we announced will be the first major phase and that there will be much more to come. I said, “This is the first 10 gigawatts. I assure you of that.”

What will the coming months look like?

Incremental deployments tied to the gigawatt milestones — as each gigawatt of capacity is completed, incremental capital and systems will be deployed.
Ongoing integration and performance tuning between OpenAI’s models and NVIDIA’s platforms to extract maximum efficiency and capability.
Cooperative buildouts with cloud and infrastructure partners across multiple geographies.
Further announcements on additional partners, timelines, and technology enhancements as the project scales.

Sam emphasized that OpenAI has three core responsibilities in the months ahead: continue great AI research, build compelling products, and manage an unprecedented infrastructure challenge. I agreed, and I added that NVIDIA will be intensely focused on delivering the systems, software, and supply chain contributions needed to make that possible.

⚖️ Governance and Strategic Alignment

One close question concerned governance: what influence do large corporate investors like NVIDIA and Microsoft have over OpenAI? Sam addressed this directly: “We're thrilled to have them both as partners. They're passive investors. Our nonprofit and board are in control.”

That matters because it preserves OpenAI’s stated mission orientation while aligning with the technical and operational strength of partners. The relationship is structured to bring strategic partners close to the operation and success of OpenAI without changing its governance commitments.

🛡️ Risks, Constraints, and How We’re Addressing Them

No project at this scale is without risk. We discussed several constraints and the mitigation paths we’re taking:

Supply chain capacity: We’re coordinating closely with suppliers and using our scale to prioritize allocations.
Power and energy demands: Data centers will need careful design, renewable integrations, and power procurement strategies to be sustainable.
Regulatory and export controls: We are aligning deployments with applicable laws and working with partners and regulators worldwide to ensure compliance.
Security and misuse: Building safe, robust systems is a cross-cutting concern that spans software, hardware, and policy.

Addressing these risks requires integration across industries and public-private collaboration — a theme we reiterated throughout the conversation.

🏥 Use Cases and Impact: From Curing Disease to Universal Education

One of the most compelling parts of the discussion was the concrete examples of what abundant compute could unlock. Sam and I discussed research-level impacts: accelerating drug discovery, enabling novel scientific insights, and building AI systems that can act proactively for users.

I framed the moral imperative this way: if society can choose to use compute for the most pressing human needs — curing diseases or delivering free education at scale — we shouldn’t have to choose between them. We need capacity so that many of these goals can be pursued in parallel.

Here are a few high-impact domains enabled by abundant, frontier compute:

Biomedical research: Simulating molecular interactions, speeding candidate drug discovery, and augmenting laboratory research with models that can hypothesize and test at scale.
Education: Personalized tutoring and curriculum delivery scaled to billions of learners.
Climate science: High-fidelity modeling and forecasting that require exascale-level compute.
Enterprise transformation: Automating repetitive knowledge work, accelerating R&D cycles, and enabling new product classes.
Creative industries: AI-assisted design, content creation, and new media formats.

These are just the start. When you combine rising model capability with massive compute availability, the range of applications expands significantly.

📊 Economic and Market Reactions

It was noticeable on the CNBC app during our conversation that NVIDIA’s stock moved to the highs of the day. Market participants are interpreting this news as strategic proof that NVIDIA remains central to the AI acceleration. The combination of a multi-year supply engagement and massive deployment plans reinforces NVIDIA’s role in the AI supply chain.

From OpenAI’s perspective, this commitment increases capacity certainty and helps to stabilize long-term planning for product development and global deployments. It also signals to other partners and customers that large-scale, durable AI infrastructure investments will be supported by robust supply pipelines and collaborative engineering.

🧩 Why This Project Is Additive — Not Contrary — to Other Announcements

We were clear that this 10 GW announcement is additive to what OpenAI and other partners have already contracted. Oracle, Microsoft, CoreWeave and others have significant commitments that are already being factored into financial planning and guidance. This project layers on top of the existing infrastructure landscape and accelerates capacity in a way that complements prior arrangements.

As I noted during the conversation: “This is additive, incremental on top of that, which just kind of puts it in perspective, the scale of AI computing that's needed for the world.”

🔁 Long-Term Vision: What the Next Decade Might Look Like

Projecting a decade ahead is inherently speculative, but the trajectory we discussed points to a few broad shifts:

Ubiquity of AI: Every interaction, image, and video will be touched, reasoned about, or generated by AI in some way.
Persistent personal agents: Many people will have always-on AI assistants that perform tasks proactively.
Compute-scarce economics: The economy will be increasingly powered by compute, making efficient resource allocation and global distribution critical.
New industries and transformations: Entirely new classes of products and services will emerge that we cannot fully predict today.

In short, the next decade will be defined by integration: AI embedded into devices, applications, enterprise workflows, and the fabric of online experiences, underpinned by hyperscale data centers and edge platforms.

🗺️ How We’ll Measure Success

Success for a project of this nature can and should be measured on multiple axes:

Operational throughput: Delivered gigawatts and the associated uptime, performance, and efficiency of deployed systems.
Model impact: The improvement in model capability per unit of compute and the rate of new breakthrough capabilities enabled.
Access and affordability: Lowering the cost of AI services to enable wide adoption.
Societal benefit: Measurable impacts on healthcare, education, scientific research, and productivity.
Safety and governance: Robust mechanisms to reduce misuse, ensure fairness, and protect privacy and security.

We will be judged not just on how many GPUs we deliver, but on the value those GPUs enable for the world.

📣 Closing Thoughts: Why I Believe This Matters

This announcement is the largest computing infrastructure commitment in history because the opportunity and demand for AI have scaled faster and further than most people anticipated. I came to the conversation with Sam and Greg because I believe NVIDIA is uniquely positioned to help deliver the compute platforms, software, and systems engineering required to make OpenAI’s ambitious vision deliverable at scale.

Jensen Huang: “This is the AI industrial revolution arriving. It's a very big deal.”

We chose to structure our investment and deployment approach around progressive, gigawatt-linked milestones because it matches the reality of how infrastructure is built: physical projects, delivered sequentially, integrated into a global supply chain. It’s deliberate, pragmatic, and designed to accelerate the availability of frontier AI capability to researchers, developers, entrepreneurs, and everyday users.

Finally, I want to emphasize the collaborative spirit behind this initiative. This is not a single-company endeavor. OpenAI’s partnerships with Microsoft, Oracle, Stargate, SoftBank, and many others are all part of a mosaic that will deliver compute where it’s needed. We are taking this first step together because the opportunities for humanity — from cures to education to climate modeling — are too big to tackle in isolation.

📬 What I’ll Be Watching and What You Can Expect Next

In the coming months, here’s what I’ll be focused on and what you can expect from us:

Updates on deployment milestones tied to each gigawatt.
Details on the Vera Rubin capacity and system-level innovations emerging from the collaboration.
Announcements of additional partners and multi-regional buildouts.
Progress on sustainability and energy procurement related to the data centers we build.
Further clarity on governance and how investments are structured to preserve OpenAI’s mission orientation.

My commitment is to share progress transparently and to work with the global community to ensure this infrastructure is used to generate broad-based benefits.

🔗 Resources and Next Steps

If you want to dig deeper, you can find the official announcement and technical blog post on the NVIDIA blog. For developers and organizations interested in preparing for the next wave of AI deployment, consider these steps:

Evaluate where your workloads fit on the edge-to-cloud spectrum and plan hybrid architectures accordingly.
Invest in software and tooling that supports model parallelism, distributed training, and efficient inference.
Engage with cloud and infrastructure partners early to understand capacity timelines and contractual terms.
Advocate for policies that attract global talent and streamline immigration processes for high-skill workers.
Prioritize responsible AI practices as compute scales up — safety, security, and fairness must be front and center.

🤝 A Final Note of Gratitude

Building the infrastructure for the next era of AI is an extraordinary undertaking that requires technological innovation, fiscal commitment, and cooperative effort across companies and nations. I am grateful to Sam Altman, Greg Brockman, our partners at Microsoft, Oracle, Stargate, SoftBank, and the countless engineers and teams working around the clock to make this possible.

We have only just begun. The 10 gigawatts we announced is a beginning — a massive, tangible start to a journey that will reshape industries and expand human capability. I am excited and humbled to be on this journey with so many brilliant partners and colleagues.

Thank you for reading. I’ll share updates as we reach the next milestones.