NVIDIA Is Building Far More Than a Chip Company

The Other 50%

One of the most important numbers in NVIDIA's latest results received surprisingly little attention. For the quarter ended July 26, 2026, NVIDIA generated approximately $89 billion of Data Center revenue. Roughly $48.7 billion came from hyperscale customers, while another $40.3 billion came from AI clouds, industrial and enterprise customers.

Management has indicated that non-hyperscalers - including sovereign customers, regional AI clouds, enterprises, edge deployments and air-gapped data centers - are approaching roughly half of the Data Center business.

That distinction is extremely important. The conventional bear case says Microsoft, Google, Amazon, Meta and other hyperscalers are NVIDIA's biggest customers, those companies are designing proprietary AI chips, and eventually they will use more of their own silicon and become less dependent on NVIDIA.

That risk is real. But NVIDIA obviously understands it too. Its response appears to be much broader than simply designing a faster GPU: NVIDIA is expanding the universe of organizations capable of becoming major consumers of AI compute.

Instead of depending indefinitely on a small collection of hyperscalers with enormous balance sheets, NVIDIA can help create thousands of additional AI-infrastructure customers. That is where the AI-factory strategy becomes particularly interesting.

From Selling GPUs to Selling AI Factories

Jensen Huang increasingly describes modern data centers not simply as data centers, but as AI factories. The distinction is important.

A traditional data center primarily stores information and runs applications. An AI factory takes electricity, data and computing capacity and turns them into economically useful intelligence: tokens, predictions, simulations, agent actions, autonomous decisions and robotic movements.

NVIDIA increasingly wants to provide the architecture around that factory: GPUs, CPUs, networking, NVLink, rack-scale systems, CUDA, software libraries, inference software, simulation, reference architectures and AI models.

The sale is gradually changing from 'Here is the chip you need' to 'Here is how you build the entire AI factory.' That is a fundamentally larger market.

Wall Street Can Finance the AI Factory

There is one enormous obstacle to expanding AI infrastructure beyond the hyperscalers: capital.

Microsoft, Amazon, Alphabet and Meta can spend tens of billions of dollars because their balance sheets allow it. A startup, university, regional AI cloud or many governments cannot simply write a multibillion-dollar check.

NVIDIA is now trying to solve that problem too. NVIDIA has partnered with major financial institutions including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR around financing platforms intended to mobilize more than $500 billion of third-party capital for AI infrastructure.

AI compute is beginning to be treated not simply as technology spending, but as an infrastructure asset capable of generating cash flow. Institutional investors already finance power plants, airports, toll roads, solar farms, telecom towers and data centers because those assets can generate predictable streams of income. An AI factory can potentially do something similar by selling compute.

That changes who can become an NVIDIA customer. A company that cannot afford a $5 billion AI factory outright might nevertheless be able to finance one. A government could finance sovereign compute. A regional cloud company could expand capacity. A university network could operate dedicated scientific infrastructure. Capital markets therefore become another mechanism for expanding NVIDIA's total addressable market.

Compute Is Becoming an Asset Class

The financialization of computing goes even further. Markets are beginning to develop pricing instruments around AI compute itself.

For institutional capital to finance something efficiently, investors need to understand expected utilization, cash flow, depreciation, residual value and future price. Once compute has transparent pricing and increasingly liquid markets, lenders can potentially model an AI factory more like a conventional infrastructure project.

That can lower the cost of capital. Lower capital costs make additional AI factories economically viable. More AI factories create additional demand for NVIDIA systems. That is the flywheel.

Falling Token Costs Could Make NVIDIA's Market Larger

Another frequent concern is that AI computing becomes dramatically cheaper. It almost certainly will as chips improve, models become more efficient, software improves, utilization rises, networking improves and inference becomes more specialized.

But falling prices do not necessarily mean NVIDIA's opportunity shrinks. They may mean the opposite. History repeatedly shows that when the cost of a transformative resource collapses, consumption can explode. Computing did this. Storage did this. Bandwidth did this. AI inference may do the same thing.

Applications that make no economic sense when inference is expensive suddenly become obvious when intelligence becomes cheap: millions of autonomous corporate agents, persistent personal assistants, continuous factory simulations, large-scale scientific experimentation, continuously optimized software, robots reasoning every second, autonomous vehicles interpreting huge sensor streams, dynamic virtual worlds and individualized AI tutoring.

NVIDIA does not necessarily need the economics of today's token to survive forever. It needs the quantity of intelligence consumed to increase faster than the cost of intelligence falls.

NVIDIA Is Moving Up Into the Model Layer

There is another defensive problem NVIDIA must consider: what happens if the largest AI-model companies eventually control their own chips? Google has TPUs. Amazon has Trainium. Microsoft has custom silicon. Meta is developing accelerators. Other frontier labs are pursuing increasingly deep infrastructure strategies.

If control of AI ultimately requires control over the entire stack, NVIDIA cannot assume customers will remain permanently dependent on NVIDIA at every layer. Its response appears increasingly clear: move up the stack too.

NVIDIA already develops substantial families of AI models. Nemotron targets agentic AI. Cosmos targets physical AI and world modeling. Isaac GR00T targets robotics. Alpamayo targets autonomous vehicles.

NVIDIA's acquisition of Hugging Face takes the company even deeper into AI-model distribution and development. The approximately $13 billion transaction gives NVIDIA exposure to one of the world's most important repositories and ecosystems for open AI models.

NVIDIA does not necessarily have to replace OpenAI or Anthropic. It needs to make NVIDIA infrastructure the easiest place on which to build, train, customize and deploy intelligence.

The Next AI Market Is Physical

The next great expansion of artificial intelligence may move beyond screens. Today's generative AI primarily generates digital output: text, images, video, code and analysis. Physical AI acts in the real world: cars, robotaxis, factories, warehouses, industrial robots, humanoid robots, delivery systems and autonomous machines.

NVIDIA has spent years preparing for this transition. Its stack increasingly encompasses the computers inside machines, the infrastructure used to train them, simulation environments, models and development software.

NVIDIA can potentially monetize physical AI at several points: hardware used for training, hardware used for inference, networking, simulation, software, foundation models and the AI-factory infrastructure behind everything. Increasingly, it can also help provide the capital necessary to finance that infrastructure.

NVIDIA Is Quietly Becoming One of the World's Largest Technology Investors

NVIDIA is no longer simply selling infrastructure to companies participating in the AI boom. It increasingly owns pieces of those companies.

As of July 26, 2026, NVIDIA disclosed approximately $99 billion of equity investments plus another $25 billion of equity-investment commitments. Only two years earlier, NVIDIA's equity portfolio was roughly $2.2 billion.

The portfolio included roughly $48 billion in publicly traded securities, $48 billion in private and other non-marketable securities, and $3 billion in equity-method investments. This is no longer a side project. It is becoming a meaningful second capital-allocation engine inside NVIDIA.

Investment Scorecard

Known entry costs versus approximate stake values. Public-market values use the researched September 9, 2026 reference prices unless otherwise noted. These are unrealized marks, not realized cash profits.

Company Investment Details

The main article stays concise. Open any company below to see the underlying investment summary and ROI logic.

NVIDIA's Public Portfolio Was Worth More Than $63 Billion

At June 30, NVIDIA's reported 13F portfolio contained eight positions worth approximately $63.4 billion in aggregate. The largest included Intel at approximately $30 billion, SpaceX at approximately $21 billion, CoreWeave at approximately $4.7 billion, Coherent at approximately $3.1 billion, Nokia at approximately $2.2 billion and Synopsys at approximately $2.15 billion, with smaller positions in Nebius and Generate Biomedicines.

Even that does not represent the entire NVIDIA investment portfolio. NVIDIA disclosed approximately $99 billion of total equity investments because tens of billions more sit in private companies, non-marketable securities and equity-method investments.

The Private Portfolio Could Become Even More Important

Recent reporting indicates NVIDIA's private portfolio includes major investments or commitments across leading AI companies and infrastructure businesses, including OpenAI, Anthropic, Safe Superintelligence, Poolside, Reflection AI and SB Energy.

A percentage ROI cannot responsibly be assigned to many of these private investments yet. Knowing that NVIDIA invested a certain amount and knowing a company's headline valuation does not reveal the exact value of NVIDIA's stake. Share class, ownership percentage, dilution, preferred terms, additional rounds, conversion rights and secondary transactions all matter.

But strategically, these private investments may be even more important than the public portfolio because they place NVIDIA inside the companies creating the next generation of AI demand.

NVIDIA Has Something Most Venture Capitalists Do Not

NVIDIA can see what is happening inside the AI economy before most outside investors can. It knows who is ordering GPUs, who is rapidly expanding clusters, which workloads are exploding, which startups suddenly need enormous amounts of compute, where power shortages are developing, which networking bottlenecks exist and which AI-cloud companies are achieving high utilization.

In many cases NVIDIA does not need to guess which companies have genuine AI demand. NVIDIA can see the demand through its own order book. That is an extraordinary informational advantage for capital allocation.

NVIDIA May Be Becoming the Berkshire Hathaway of AI

Calling NVIDIA the 'Berkshire Hathaway of technology' or 'Berkshire Hathaway of AI' is not literally accurate. Berkshire's primary business is owning and allocating capital across businesses, while NVIDIA's primary business remains computing.

But the comparison captures something important. NVIDIA increasingly has two different ways to capture the economic value created by the AI revolution: first, sell the infrastructure - GPUs, networking, systems, software, AI factories, robotics and autonomous-driving infrastructure; second, own pieces of the ecosystem - AI labs, cloud companies, semiconductor companies, networking companies, optics companies, suppliers and customers.

If those companies grow because the AI economy grows, NVIDIA can participate through both operating earnings and investment appreciation.

The Strategy Can Reinforce the Ecosystem It Owns

NVIDIA can sometimes do something a conventional holding company cannot: actively help make its portfolio companies more valuable through technology, customer relationships and ecosystem development.

CoreWeave is a useful example. NVIDIA supplies the hardware that powers its cloud, invests in CoreWeave, CoreWeave raises additional capital, constructs more AI factories and purchases more NVIDIA systems. If CoreWeave's revenue grows, its equity may appreciate. NVIDIA potentially earns both GPU revenue and investment returns.

The same concept applies differently across the ecosystem. Investing in Nokia can accelerate AI networking. Investing in Coherent can help expand optical manufacturing capacity. Investing in AI labs can stimulate workloads that require enormous amounts of NVIDIA compute. This is less like ordinary venture investing and more like ecosystem capital allocation.

The Important Counterargument: Circular Financing

Critics have raised an understandable objection: if NVIDIA invests in companies that subsequently purchase NVIDIA hardware, aren't NVIDIA's investments artificially creating its own revenue? That concern should not simply be dismissed.

The important distinction is between fake economics and ecosystem financing. If NVIDIA gives a company money, the company merely buys NVIDIA GPUs, there are no external customers and the infrastructure has no economic value, the concern would be serious.

But if NVIDIA invests alongside outside capital, the company constructs productive infrastructure, external customers rent it, the infrastructure generates revenue, the company grows and its equity appreciates, that is a different economic model.

The question investors should ask is not simply 'Is the financing circular?' It is: 'Is real external demand being created at the end of the circle?'

NVIDIA Is Building Its Own Economic Ecosystem

Apple created an ecosystem around the iPhone. Microsoft created one around Windows. Amazon created one around AWS. Google created one around search, Android and cloud. NVIDIA appears to be creating one around accelerated intelligence.

Inside that ecosystem are chip manufacturers, networking suppliers, cloud providers, frontier AI labs, open models, robotics companies, autonomous vehicles, energy providers, data centers, financial institutions, governments, developers and enterprises.

NVIDIA supplies technology to many of them, invests in many of them and increasingly helps finance them. Some then become large NVIDIA customers. That creates an economic ecosystem where NVIDIA can participate in extraordinary amounts of value creation.

What NVIDIA May Ultimately Be Building

Imagine a government, enterprise, university consortium or regional AI company deciding in the future that it wants its own large-scale artificial-intelligence capability.

NVIDIA's potential offering could increasingly resemble: the chips, CPUs, networking, servers, rack architecture, operating software, AI models, simulation environment, robotics platform, autonomous-driving platform, data-center blueprint, developer ecosystem and financing.

The customer gets something approaching a vertically integrated AI factory without needing to become Microsoft, Google or Amazon. NVIDIA sits somewhere inside almost every layer.

The Risk to the Thesis

None of this makes NVIDIA invulnerable. Its largest customers are spending enormous sums developing alternative chips. AI infrastructure spending could overshoot demand. Competitors could attack CUDA. Custom accelerators could become more efficient for specialized workloads. Governments could challenge NVIDIA's market power. Financing arrangements could expose NVIDIA to credit and residual-value risk. Investment losses could become substantial. Customer financing creates legitimate circularity concerns. And NVIDIA's involvement across multiple layers could attract increasing antitrust attention.

Not every NVIDIA investment will work. Synopsys already demonstrates that the portfolio does not simply move upward. A portfolio that increased from roughly $2 billion to almost $100 billion in two years introduces an entirely new category of capital-allocation risk that needs to be watched closely.

But NVIDIA No Longer Needs Every Dollar of Value to Come From GPU Margin

The NVIDIA bear case often assumes today's extraordinary GPU margins eventually fall. That is probably correct. Competition will increase, custom chips will improve and hardware will commoditize to some degree.

But the total NVIDIA thesis does not require today's margins to persist forever. NVIDIA can potentially capture value through hardware, networking, software, enterprise AI, models, AI factories, inference, robotics, autonomous vehicles, infrastructure financing and equity ownership across the AI ecosystem.

The relevant question may therefore no longer be 'How long can NVIDIA maintain today's GPU margins?' It may increasingly become: 'How much of the global AI economy can NVIDIA participate in?' Those are radically different questions.

The Bigger NVIDIA Thesis

NVIDIA's greatest asset may no longer simply be the GPU. It is the ecosystem forming around the GPU.

The company controls the dominant accelerated-computing architecture, one of the world's most entrenched AI software platforms, a massive developer ecosystem, rapidly expanding AI-model families, autonomous-driving infrastructure, robotics infrastructure and increasingly important relationships with major financial institutions.

And now NVIDIA has built an equity portfolio approaching $100 billion. The portfolio includes a roughly $21 billion SpaceX position at June 30; a $5 billion Intel investment that was still worth roughly $22.8 billion by September 9; a $1 billion Nokia investment worth roughly $1.79 billion; a $2 billion Coherent investment worth roughly $2.36 billion; billions more in CoreWeave; and tens of billions invested across private AI companies.

NVIDIA is not simply investing cash that happens to be sitting on its balance sheet. Many of its investments appear designed to expand the economic ecosystem surrounding NVIDIA itself.

If an AI cloud succeeds, NVIDIA may sell it GPUs and own equity in it. If an optical supplier succeeds, NVIDIA may obtain more infrastructure capacity and own equity in it. If an AI lab succeeds, NVIDIA may sell it enormous quantities of compute and participate in its valuation. If an infrastructure provider expands, NVIDIA may sell the hardware inside the facility while owning part of the business operating it.

The hyperscalers will continue building proprietary chips. NVIDIA knows that. So NVIDIA appears to be preparing for a world in which AI becomes much larger than the hyperscalers themselves.

NVIDIA would no longer simply sell the picks and shovels for the AI gold rush. It would help design the mine, supply the machinery, finance its construction, provide the operating system, invest in the miners - and own pieces of the companies that discover the gold.

That is a much bigger company than a chipmaker. And it may be the NVIDIA Jensen Huang has been building all along.