Lisa Su is looking at the artificial-intelligence boom from a position that most semiconductor executives would love to have.

Her problem is that even after AMD has significantly expanded production, customers still want more chips than the company can supply.

That was the message from AMD's CEO in Taiwan on October 6, where she said demand for AI computing remains “very high” and is expected to stay elevated for several years. AMD plans to substantially increase chip supply in 2027, but Su made clear that simply adding capacity may not be enough.

The comments are important because AMD is no longer a distant challenger in the AI semiconductor market.

Its market value recently crossed $1 trillion, putting it among the world's largest technology companies.

And yet its biggest problem is apparently not demand.

It is finding enough manufacturing capacity to keep up with demand.

In an ordinary semiconductor cycle, an executive complaining that supply cannot keep pace with demand would be welcome news.

But at AMD's current scale, the implications are much larger.

Su said AMD has already increased supply during 2026, but demand remains higher than what the company can provide. The company is therefore preparing to substantially increase supply in 2027 and is working with major Taiwanese partners to expand production across CPUs and GPUs.

That means AMD's challenge is increasingly about execution.

The company needs enough advanced wafers.

It needs enough memory.

It needs packaging and assembly capacity.

And it needs a supply chain capable of moving enormous quantities of high-value components through an industry where demand can change much faster than semiconductor factories can.

Su's visit to Taiwan was therefore more than a routine executive trip.

It was a capacity-planning mission.

One of Su's main objectives was to discuss supply-chain expansion with Taiwanese partners, including Foxconn and Taiwan Semiconductor Manufacturing Co.

AMD depends heavily on TSMC for advanced semiconductor manufacturing, making the foundry's ability to add capacity central to AMD's growth plans. Su said AMD needs additional advanced wafer capacity and is now planning roughly three to five years ahead because of the strength of anticipated demand.

That long planning horizon is one of the clearest signals of how different the AI semiconductor cycle is from traditional electronics demand.

Smartphone companies can change product plans within months.

AI infrastructure operators make multiyear commitments.

A hyperscale data center can require billions of dollars of investment before it becomes operational.

That gives chipmakers much more visibility into future demand—but it also means they have to secure manufacturing capacity years in advance.

AMD is trying to get ahead of that curve.

AMD's supply problem does not end at the wafer.

Su said the company is also working closely with memory manufacturers to make sure adequate supplies are available. She was scheduled to visit South Korea to meet with companies including Samsung Electronics and SK Hynix, two of the world's biggest memory-chip manufacturers.

That is significant because high-performance AI systems rely heavily on advanced memory.

The latest generation of AI accelerators can process enormous amounts of data, but their performance depends on the ability to move that data quickly.

High-bandwidth memory, or HBM, has therefore become one of the most important components in the AI hardware stack.

Micron recently reported extraordinary demand for AI-related memory and said its long-term customer commitments had climbed sharply, with major customers securing future capacity. The company forecast $61.5 billion in quarterly revenue for its fiscal first quarter, far ahead of Wall Street estimates.

AMD is operating inside that same supply squeeze.

It can design the accelerator.

TSMC can manufacture it.

But without enough HBM and packaging capacity, the finished AI system cannot ship.

This is one of the most important changes in the semiconductor industry.

At the beginning of the AI boom, the story was primarily technological.

Which company has the strongest model?

Which accelerator is fastest?

Which software ecosystem is easiest to program?

Now the bottleneck is increasingly physical.

Who has enough wafers?

Who can secure advanced packaging?

Who gets priority from memory suppliers?

Who can build enough data centers and connect them to electricity?

The companies that solve those problems can potentially capture a disproportionate share of the AI economy.

That's why AMD is spending so much executive attention on its supply chain.

The company's ability to sell more AI accelerators is constrained not by a lack of customers, but by the industry's ability to manufacture enough of them.

AMD's opportunity should not be confused with Nvidia's position.

Nvidia remains the dominant supplier of AI accelerators and benefits from a deeply entrenched software ecosystem.

But AMD has been steadily expanding its role as an alternative supplier.

Its Instinct accelerators have gained traction among major cloud and technology customers, while the company has also been investing aggressively in software and system-level capabilities.

AMD's strategy increasingly resembles a broader challenge to Nvidia rather than a narrow attempt to sell cheaper GPUs.

The company's recent acquisition of AI startup World Labs for $8.2 billion is part of that effort. World Labs, co-founded by AI pioneer Fei-Fei Li, develops “world models” designed to help AI systems understand physical environments and could strengthen AMD's position in robotics and physical AI.

AMD is trying to build an ecosystem.

And the more complete that ecosystem becomes, the more important its available computing capacity becomes.

Crossing a $1 trillion market value is not merely a symbolic achievement.

It reflects investors' growing belief that AMD has become a durable participant in the AI infrastructure market.

But it also creates higher expectations.

Once a company reaches that scale, shareholders begin looking for enormous revenue growth, expanding margins and increasingly reliable cash flow.

A supply shortage can help the story.

It suggests customers want more.

But it can also create frustration if AMD cannot convert that demand into actual shipments.

That's why 2027 matters so much.

The company is telling investors that supply will increase substantially.

The market will eventually judge how much additional revenue and profit that capacity creates.

AMD's warning comes at a time when almost every major part of the AI hardware ecosystem is discussing capacity constraints.

Micron is adding memory capacity.

TSMC is expanding advanced-wafer manufacturing.

Foxconn is reporting extraordinary demand for AI servers and networking equipment.

Broadcom is building financing structures around major AI customers.

The common theme is that AI capital expenditure is still running ahead of what the physical supply chain can comfortably accommodate.

That is powerful evidence that the AI infrastructure cycle remains alive.

But it also reveals why the industry needs enormous capital spending.

The next generation of AI cannot simply run on the hardware already installed.

New data centers need new accelerators.

New accelerators need new memory.

New memory requires new factories.

And those factories require billions of dollars.

The system is feeding itself.

Interestingly, Su's comments were not entirely bullish.

She also urged AI companies to cooperate on safety.

She said AI can be “incredibly good for the world” but argued that companies need to work together to make the technology both capable and safe. Her remarks came shortly after a meeting between President Donald Trump and major technology executives at which companies agreed to work with independent auditors to evaluate whether AI systems operate as intended.

That introduces another dimension to the semiconductor race.

The industry is not only building more powerful systems.

It is also beginning to face questions about how those systems should be deployed.

More capable AI means more computing demand.

But greater capability also means greater pressure for testing, monitoring and safeguards.

The hardware race and the governance debate are therefore developing simultaneously.

The strongest part of Su's message is that demand appears to be stronger than supply.

That's exactly the problem shareholders want AMD to have.

But the company now has to execute.

It has to secure TSMC capacity.

It needs enough advanced packaging.

It needs memory supplies from Samsung and SK Hynix.

It needs customers to keep committing capital.

And it needs AI accelerator revenue to grow fast enough to justify the enormous valuation investors have assigned to the company.

The broader AI market is still expanding rapidly.

Micron's results, Foxconn's server growth and Nvidia's continuing infrastructure demand all point in the same direction.

AMD is now trying to capture a much larger share of that expansion.

Lisa Su's message from Taiwan was therefore unusually revealing.

AMD isn't preparing for an AI slowdown.

It is preparing for a world in which the biggest problem may be finding enough chips to satisfy everyone who wants them.

The next phase of the AI race may be won not only by the company with the best processor.

It may be won by the company that can manufacture enough of them.