AI Data Centre Boom Drives GlobalFoundries and Marvell to Expand Chip Production
The artificial-intelligence infrastructure race is creating demand for more than powerful processors.
It is also creating an enormous need to move data between those processors quickly enough to keep them productive.
That challenge is driving a new expansion in semiconductor manufacturing.
GlobalFoundries and Marvell Technology announced on September 17 that they are expanding their collaboration to increase manufacturing capacity for semiconductor technology used in high-speed optical connectivity for AI data centres.
The agreement highlights an increasingly important part of the AI economy.
Building faster AI chips is only one part of the problem.
Those chips also have to communicate.
As AI systems grow larger, the infrastructure connecting processors, memory and data-centre equipment can become a major performance bottleneck.
AI Is Changing the Architecture of Data Centres
Traditional data centres were built to support websites, databases, cloud applications and corporate computing.
AI workloads can be dramatically more demanding.
Training a large artificial-intelligence model can involve enormous clusters of specialised processors operating simultaneously.
Rather than treating each server as an isolated machine, modern AI infrastructure increasingly behaves like one gigantic computing system.
That requires data to travel between components at extremely high speeds.
If information cannot move quickly enough, expensive processors can spend valuable time waiting.
That makes connectivity strategically important.
The Network Is Becoming Part of the AI Race
Much of the public discussion around artificial intelligence has focused on graphics processing units and other accelerators.
Those processors remain critical.
But AI infrastructure contains an entire ecosystem around them.
It needs memory.
It needs networking equipment.
It needs optical components.
It needs power-management technology.
It needs cooling systems.
And it needs semiconductor components capable of connecting all those systems efficiently.
As data centres become larger, communication between computing components becomes increasingly difficult to handle using conventional electrical connections alone.
Optical technology provides another option.
Why Optical Connectivity Matters
Light can carry enormous amounts of information.
That makes optical communication particularly useful when large quantities of data need to move rapidly across a data centre.
Fibre-optic connections are already fundamental to telecommunications networks.
AI is increasing their importance inside computing infrastructure itself.
A large AI cluster may contain thousands or even tens of thousands of processors.
Those processors constantly exchange data.
Increasing the speed and efficiency of those connections can therefore improve the performance of the entire system.
This is where specialised semiconductor technology becomes important.
Chips involved in optical communication help translate, process and manage the signals carrying information through these networks.
GlobalFoundries Brings Manufacturing Scale
GlobalFoundries is one of the world's major contract semiconductor manufacturers.
Unlike companies that design chips but outsource production, foundries operate the factories where semiconductor designs become physical products.
These factories are extraordinarily complex.
Producing chips requires highly controlled environments, specialised machinery and manufacturing processes capable of working at microscopic scales.
Building additional semiconductor capacity therefore takes substantial capital and planning.
For customers such as Marvell, having manufacturing capacity secured can be strategically important when demand is rising quickly.
The expanded collaboration is designed to increase capacity for technology supporting AI data-centre connectivity.
Marvell Focuses on the Connections Between Computers
Marvell operates across several semiconductor markets but has become increasingly exposed to data-centre infrastructure.
The company develops technology used in networking and connectivity.
That places it in an interesting position within the AI boom.
Instead of competing only for the processor doing the artificial-intelligence calculations, Marvell can benefit from the infrastructure surrounding those processors.
Think of an AI data centre as a city.
Powerful AI accelerators may be the buildings where work happens.
But those buildings become much less useful without roads connecting them.
Networking technology provides those roads.
And as the city becomes larger, the roads have to become faster.
AI Infrastructure Is Becoming a Supply-Chain Story
Artificial intelligence is often discussed as software.
Chatbots, image generators and coding assistants are the products consumers see.
Underneath those applications, however, sits an enormous physical economy.
AI requires semiconductor fabrication plants.
It requires servers.
It requires fibre.
It requires cooling equipment.
It requires electrical infrastructure.
It requires data-centre buildings.
And increasingly, it requires specialised connectivity technology.
This is why the economic impact of AI is spreading far beyond companies developing models.
A business may never create an AI chatbot and still benefit significantly from AI investment if it supplies something essential to the infrastructure.
Investors Noticed the Announcement
Financial markets responded positively to the expanded partnership.
GlobalFoundries shares rose around 4%, while Marvell shares gained about 6.3% in morning trading following the announcement.
A single day's share-price movement should not be interpreted as proof that a strategy will succeed.
But the reaction illustrates how closely investors are watching companies positioned around AI infrastructure.
The market increasingly distinguishes between businesses that simply mention artificial intelligence and those supplying physical components required to build it.
Manufacturing capacity falls firmly into the second category.
The AI Chip Shortage Is Broader Than GPUs
When people hear about AI semiconductor shortages, they may immediately think about the most advanced accelerators.
But a data centre cannot operate on accelerators alone.
Hundreds of supporting components are needed.
If even a relatively inexpensive component becomes unavailable, production of a much more expensive system can be delayed.
This is a familiar lesson from the automotive semiconductor shortage earlier in the decade.
A vehicle containing thousands of dollars of electronics could sit unfinished because one small chip was missing.
AI infrastructure creates similar supply-chain dependencies.
That makes capacity planning increasingly important.
Faster Models Require Faster Infrastructure
Artificial-intelligence models continue to increase in complexity.
At the same time, companies are trying to make AI responses faster and serve more users.
Both trends increase infrastructure requirements.
Training is demanding.
Running AI models at scale can be demanding too.
When millions of people interact with AI services, data centres have to process huge numbers of requests continuously.
That increases pressure on the entire computing system.
Improving networking performance can help ensure processors are used more efficiently.
In a world where advanced AI hardware can be extremely expensive, even small efficiency improvements can have meaningful financial value.
Energy Efficiency Is Becoming Important Too
Moving data consumes electricity.
At enormous scale, that matters.
AI data centres are already attracting attention because of their electricity requirements.
Technology that transfers information more efficiently could therefore have value beyond raw performance.
If optical connectivity reduces the energy required to move data over certain distances, it can contribute to better overall data-centre efficiency.
That does not solve the industry's energy challenge.
But improving efficiency across processors, networking, cooling and power systems will be necessary if AI infrastructure continues expanding.
Semiconductor Partnerships Are Becoming More Strategic
Chip companies historically competed heavily on design.
Increasingly, manufacturing relationships themselves have become strategic assets.
Geopolitical tensions, pandemic-era disruptions and rapidly growing AI demand have demonstrated the risks of depending on limited production capacity.
Technology companies therefore want greater visibility into where their chips will be manufactured and how much capacity will be available.
Foundries benefit from longer-term commitments.
Chip designers benefit from more predictable supply.
The expanded GlobalFoundries-Marvell relationship fits that pattern.
AI Spending Is Moving Deeper Into the Supply Chain
The first phase of the generative-AI investment boom focused heavily on securing computing accelerators.
The next phase is becoming broader.
Companies are investing in networking.
Cloud providers are constructing enormous data-centre campuses.
Utilities are planning new electricity generation.
Manufacturers are developing advanced cooling systems.
Semiconductor companies are increasing production of connectivity components.
The result is an AI investment cycle that increasingly touches traditional industrial sectors.
This matters because it changes the economic story surrounding artificial intelligence.
AI is no longer only a software-sector investment trend.
It is becoming a capital-infrastructure cycle.
Connectivity Could Become a Competitive Advantage
If two AI data centres contain similar processors but one can move information between them more efficiently, their overall performance may differ.
That means networking architecture can become a source of competitive advantage.
Cloud companies are already designing increasingly customised infrastructure around their AI workloads.
The suppliers capable of improving communication between processors could therefore occupy valuable positions within the technology stack.
This is particularly important as AI clusters continue scaling.
Connecting hundreds of processors is difficult.
Connecting tens of thousands is considerably harder.
The AI Race Is About More Than the Fastest Chip
The GlobalFoundries-Marvell announcement illustrates how the AI hardware race is evolving.
The industry's attention may remain focused on headline-grabbing processors.
But the infrastructure supporting those processors is becoming just as important.
GlobalFoundries and Marvell are expanding their collaboration specifically to increase manufacturing capacity for semiconductors supporting high-speed optical connectivity in AI-powered data centres.
That may sound like a highly specialised corner of the semiconductor market.
In reality, it addresses one of the biggest engineering challenges created by modern artificial intelligence:
How do you make thousands of extremely powerful computers behave like one?
The answer requires more than faster processors.
It requires faster connections between them.
And as the AI infrastructure race accelerates, the companies building those connections are moving closer to the centre of the technology economy.