Jensen Huang is pushing Nvidia far beyond GPUs, building a business designed to profit from custom chips, networking, cloud capacity and even the financing of the AI infrastructure boom.

Nvidia's greatest threat is becoming increasingly obvious.

It is not simply AMD.

It is not Intel.

It is not even the custom artificial-intelligence chips being developed by Amazon, Google and other technology giants.

The bigger threat is the possibility that Nvidia's customers eventually build enough of their own silicon to reduce their dependence on the company's flagship GPUs.

Nvidia's answer is remarkably ambitious:

Make money from those custom chips anyway.

That is the strategic message behind the company's latest moves, including its $3.5 billion investment in Taiwan's MediaTek, which can help customers develop custom AI processors capable of connecting into Nvidia's broader data-center architecture. At the same time, Anthropic has committed roughly $35 billion to rent Nvidia-powered computing capacity from specialized cloud provider Lambda.

The two transactions look completely different on a balance sheet.

One involves Nvidia putting capital into a chip-design partner.

The other involves an AI company committing capital to consume computing capacity.

But strategically, they point in the same direction.

Nvidia increasingly wants to become the architecture of AI, not merely the company that sells the most popular AI accelerator.

That could be an even more powerful business.

Nvidia's rise was built on GPUs.

The company's chips became the standard hardware for training and running the largest AI models, creating an extraordinary demand cycle.

But the AI infrastructure being built around those GPUs is vastly more complicated than a box of processors.

Large AI systems require:

High-speed networking.

CPUs.

Memory.

Interconnects.

Cooling.

Power infrastructure.

Software.

Data-center systems.

Financing.

Nvidia is increasingly participating in almost every layer.

The company's latest strategy suggests management has recognized a simple principle:

The more infrastructure Nvidia controls, the less vulnerable it becomes to any single chip competitor.

That is a major strategic shift.

For years, Wall Street treated custom AI silicon as a direct challenge to Nvidia.

The logic was obvious.

If Amazon can build a powerful AI accelerator for AWS, it has less reason to buy Nvidia GPUs.

If Google can use its own Tensor Processing Units, Nvidia could lose a customer.

If Microsoft and Meta develop their own hardware, Nvidia's market share could eventually decline.

But Jensen Huang is taking a different approach.

Nvidia's NVLink Fusion technology allows customers to integrate custom chips into Nvidia's broader architecture.

That means a company can design its own processor while still relying on Nvidia connectivity and infrastructure.

The result is potentially clever.

Nvidia no longer needs to win every processor battle.

It needs to remain essential to the data center.

Nvidia's $3.5 billion investment in MediaTek is particularly revealing.

MediaTek is a major Taiwanese chip designer with experience building customized silicon for large customers.

By investing in the company, Nvidia is effectively strengthening the ecosystem that can produce alternative AI processors while creating more opportunities for those processors to plug into Nvidia infrastructure.

That can turn a competitive threat into a potential source of revenue.

Suppose a hyperscaler develops a custom accelerator.

Nvidia may lose the accelerator sale.

But if the custom chip uses Nvidia networking, interconnects and systems architecture, Nvidia still participates.

That is a much more resilient model.

The company's own description of an AI data center is revealing.

Jensen Huang recently characterized an AI factory in simple terms: electricity and data come in, and tokens come out.

The implication is profound.

An AI data center is becoming an industrial facility.

It has inputs.

It has computing machinery.

It has infrastructure.

It has financial economics.

And it produces a valuable output: AI inference and computation.

Nvidia increasingly wants to supply the machinery for that entire process.

That is why the company's future opportunity is becoming much larger than the GPU market alone.

One of the clearest indicators of the strategy came from Nvidia's latest earnings call.

The company said its revenue opportunity for each gigawatt of AI-factory power capacity has risen from roughly $18 billion with Hopper to about $25 billion with Grace Blackwell and roughly $40 billion with Vera Rubin.

That is an extraordinary progression.

It shows how Nvidia is increasing the amount of infrastructure it captures from each AI deployment.

Instead of simply selling processors, the company is selling more of the system around them.

That has a major implication for future revenue.

If global AI power capacity expands dramatically, Nvidia's potential sales can grow even faster if its share of each facility rises.

This strategy is already visible in networking.

AI clusters increasingly require enormous amounts of data movement between processors.

Thousands of accelerators need to communicate at extremely high speeds.

A slow network can leave expensive chips underutilized.

Nvidia's networking products therefore become essential complements to its processors.

As AI clusters grow larger, networking can represent an increasingly significant portion of the data-center budget.

That gives Nvidia another source of growth.

And critically, the demand remains relevant even in a world where customers deploy more custom chips.

Amazon Web Services is already investing heavily in proprietary AI accelerators, including its Trainium family.

Yet AWS is also planning to add approximately 2 million Nvidia GPUs while integrating its own Trainium technology more deeply into Nvidia's broader infrastructure.

This is exactly the future Nvidia wants.

A customer can use its own chip.

A customer can use Nvidia chips.

Or, increasingly, it can use both.

Nvidia can participate regardless.

That makes the company's business model less binary.

Then there is Anthropic.

The company has committed approximately $35 billion to Lambda's Nvidia-powered computing capacity, according to the Yahoo Finance report.

Again, Nvidia benefits from more than the chip sale.

The hardware becomes part of a computing service.

That computing service generates revenue for a cloud provider.

The resulting demand supports additional infrastructure investment.

The cycle ultimately drives more Nvidia deployments.

This is the difference between being a component supplier and becoming an infrastructure platform.

Nvidia is also moving closer to the financial side of the AI economy.

Building an AI data center requires enormous upfront capital.

A company needs billions of dollars before it can generate meaningful revenue.

Traditional lenders can be cautious because customer demand may be difficult to predict and GPU technology evolves rapidly.

Nvidia has increasingly responded by helping mobilize financing.

The company has worked with major financial institutions including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR around efforts designed to mobilize more than $500 billion for AI infrastructure.

That changes Nvidia's role again.

It is no longer simply making the equipment.

It is helping ensure customers can afford the equipment.

A traditional hardware manufacturer waits for the customer to find the money.

Nvidia increasingly wants to help the customer find it.

That can accelerate deployments.

More deployments mean more Nvidia products sold.

And larger infrastructure projects create long-term demand for upgrades, networking and new generations of chips.

The strategy therefore attacks another major AI bottleneck.

First the problem was chips.

Then electricity.

Now capital.

Nvidia is positioning itself across all three.

The broader Nvidia becomes, the more risks it takes on.

If Nvidia finances or supports a cloud provider, it may become indirectly exposed to that provider's economics.

If AI demand slows, some customers could struggle to make payments.

If custom chips become very competitive, Nvidia could lose processor share even while retaining architecture revenue.

If regulators object to Nvidia's influence across the AI stack, the company could face antitrust scrutiny.

The company's growing importance therefore creates both strength and vulnerability.

Nvidia has already dramatically outperformed the broader semiconductor industry.

Over the past five years, the stock has gained more than 870%, compared with around 230% for a benchmark chip index.

That performance reflects extraordinary financial growth.

But it also demonstrates why Nvidia cannot simply repeat the same strategy indefinitely.

Once a company reaches enormous scale, growth must come from new markets.

Nvidia's answer is clear:

Expand the definition of what Nvidia sells.

This may ultimately be the most important strategic shift.

The traditional semiconductor industry thinks in units.

How many chips did Nvidia sell?

What was the average selling price?

What was gross margin?

The AI infrastructure industry increasingly thinks in capacity.

How many gigawatts?

How many accelerators?

How many tokens per second?

How much computing capacity?

Nvidia is positioning itself around those larger metrics.

If one gigawatt of AI capacity represents tens of billions of dollars of potential Nvidia revenue, power infrastructure becomes another way of measuring Nvidia's addressable market.

That is a much bigger opportunity.

AMD remains a serious competitor in AI accelerators.

But if Nvidia becomes the dominant networking and systems provider, competing with its GPU alone is no longer enough.

A rival would have to compete across the entire architecture.

That is much harder.

Customers are increasingly buying integrated systems because the cost of engineering massive AI clusters is itself enormous.

A pre-integrated Nvidia architecture can save time and reduce operational risk.

That creates a form of competitive moat beyond raw chip performance.

There is another reason Nvidia can pursue this architecture strategy.

Its software ecosystem remains extremely important.

CUDA allows developers to build around Nvidia hardware and creates switching costs.

A custom chip that integrates into Nvidia infrastructure can still benefit from parts of that broader ecosystem.

That makes Nvidia's platform difficult to displace completely.

Nvidia has not disclosed exactly how much revenue it expects to generate from every custom-chip factory.

That makes the economics of NVLink Fusion difficult to quantify.

But strategically, the direction is clear.

If Nvidia can turn competing processors into complementary infrastructure components, it expands its addressable market without needing to maintain 100% control over compute.

That is a powerful hedge.

The old Nvidia story was:

AI needs GPUs.

Nvidia makes the best GPUs.

Nvidia wins.

The emerging story is:

AI needs computing factories.

Nvidia provides much of the architecture required to build and operate those factories.

That is a considerably larger idea.

It includes hardware.

Networking.

Systems.

Software.

Cloud relationships.

Financing.

And potentially custom-chip integration.

The company is effectively betting that artificial intelligence will require an infrastructure buildout comparable to previous industrial transformations.

If that happens, selling one component of the factory will not be enough.

The biggest economic opportunity will come from owning critical parts of the entire system.

Nvidia is trying to occupy that position.

It does not need every processor to bear the Nvidia logo.

It does not necessarily need every AI company to buy its own data center.

It needs to remain embedded in the architecture that allows those systems to operate.

That's the genius behind the company's latest strategy.

The industry's biggest threats are being turned into potential customers.

Custom chips can connect through Nvidia.

Cloud providers can rent Nvidia systems.

AI companies can consume Nvidia-powered capacity.

Banks can finance Nvidia-powered infrastructure.

And data centers can use an expanding range of Nvidia hardware around the central accelerator.

The result is a company moving beyond the semiconductor category itself.

Nvidia is trying to become the infrastructure layer beneath the AI economy.

And if Jensen Huang succeeds, the future may not be remembered as the era when Nvidia sold the world's best AI chips.

It may be remembered as the era when Nvidia became the company that built the machine those chips—and their competitors—ran inside.