Inside the AI Hardware Boom: AMD's Record Quarter and Marvell's $115 Billion Bet on the Future of AI

Artificial intelligence is no longer just about building better chatbots, image generators or language models. Behind these applications is a growing infrastructure industry that requires powerful processors, specialist chips, high-speed networks and enormous data centre capacity.

Two developments in November 2026 highlight how quickly this industry is changing. AMD reported strong third-quarter revenue, driven by growing demand for its data centre products. Meanwhile, Marvell Technology announced a proposed $115 billion all-stock acquisition of optical technology company Lumentum.

Although the announcements concern different parts of the semiconductor industry, they point to a common trend: the next phase of AI will depend not only on computing power but also on how efficiently data moves between processors.

AMD's Data Centre Business Drives Revenue Growth

AMD reported third-quarter 2026 revenue of $13.38 billion, a 45% increase from the same period last year. According to the company's November 3 earnings release, its Data Centre segment recorded revenue of $8.5 billion, up 97% year on year. AMD attributed the growth to strong demand for its EPYC processors and the launch of its Instinct MI450 series GPUs.

AMD chief executive Lisa Su described it as a record quarter, saying that demand for the company's Instinct and EPYC products had continued to strengthen. The company expects fourth-quarter revenue to reach approximately $17 billion, plus or minus 3%. AMD said the outlook reflected expectations of accelerating revenue and earnings growth, supported by demand across its Data Centre portfolio.

The results underline the growing importance of AI infrastructure to AMD's business.

Why AMD's EPYC, Instinct and Helios Matter

AMD's AI strategy involves more than one type of chip. Its EPYC processors handle general computing tasks in servers and data centres, while Instinct GPUs are designed for demanding artificial intelligence and high-performance computing workloads.

Training and running large AI models require thousands of calculations to take place simultaneously. Specialist GPUs help manage these workloads, while CPUs, memory and networking systems ensure that the entire infrastructure works efficiently.

AMD is also expanding its Helios rack-scale AI platform, which brings processors, accelerators and networking components together into an integrated system. In its November 3 earnings statement, Lisa Su identified the scaling of Helios and Venice production as part of the company's plans to accelerate growth. This reflects a wider shift in the industry towards building complete AI computing systems rather than competing on individual chip performance alone.

Why Marvell Is Betting $115 Billion on Lumentum

While AMD is expanding its computing capabilities, Marvell is strengthening its position in the technology used to connect AI infrastructure. On November 2, Marvell and Lumentum announced a definitive agreement under which Marvell would acquire Lumentum in an all-stock transaction.

According to their joint announcement, the deal has an implied enterprise value of $115 billion, while Lumentum's equity is valued at approximately $102.5 billion. Lumentum shareholders will receive 4.5 Marvell shares for each share they own. Following the proposed acquisition, Marvell shareholders are expected to own approximately 67% of the combined company and Lumentum shareholders 33%.

The companies expect the transaction to close in the first half of 2028, subject to shareholder and regulatory approvals. The strategic reasoning behind the deal is linked to the growing demand for faster data transmission. As AI data centres expand, thousands of processors and accelerators must exchange enormous amounts of information. Even powerful chips can become less effective if the networks connecting them cannot transfer data quickly enough.

This makes connectivity an increasingly important part of AI infrastructure. Reuters, in its November 2 report on the acquisition, said the deal would help Marvell compete more directly with Broadcom in the AI connectivity market.

Why Optical Connectivity Matters

Traditional data centre networks rely heavily on electrical signals transmitted through copper connections. However, as bandwidth requirements increase, copper-based connections face limitations involving signal loss, power consumption and transmission distances. Optical technology offers an alternative by transmitting information through light.

It can support high-speed communication over longer distances and improve power efficiency in certain applications. This makes it increasingly important as AI data centres become larger and more complex. One technology attracting attention is co-packaged optics (CPO). It brings optical components closer to the main switching chips, reducing the distance electrical signals need to travel.

The aim is to increase bandwidth while improving energy efficiency.

The Marvell-Lumentum combination is intended to bring together complementary technologies across this area. Marvell contributes expertise in networking, custom silicon and electrical and optical interconnects, while Lumentum brings high-performance lasers, optical transceivers and specialised manufacturing capabilities.

In their November 2 announcement, the companies said the combination would help develop more integrated optical solutions for next-generation AI infrastructure.

They also projected that the combined company's total addressable market could reach approximately $550 billion, with the potential for more than $1 billion in annual run-rate cost synergies. These are company projections, not guaranteed financial outcomes. Their achievement will depend on the completion and integration of the transaction and future market demand.

AMD, Marvell, NVIDIA and Broadcom: The Competition Is Expanding

The competition for AI infrastructure is no longer limited to companies developing graphics processors. NVIDIA remains a major player in AI accelerators and networking systems. AMD is expanding through its EPYC processors, Instinct GPUs and integrated computing platforms.

Marvell is focusing on custom silicon and connectivity, while Broadcom remains a significant competitor in custom AI chips and networking technology. According to a Reuters report, the Lumentum acquisition would strengthen Marvell's ability to challenge Broadcom in the AI infrastructure market. The news agency also reported that NVIDIA had invested $2 billion each in Marvell and Lumentum earlier in 2026.

The changing competitive landscape shows that computing power alone will not determine the next phase of AI development. Networking, data transmission, integration and energy efficiency are becoming equally important.

The Risks Behind the AI Infrastructure Boom

Despite the growth opportunities, expanding AI infrastructure comes with significant costs.

Building large data centres requires substantial investment in advanced chips, electricity, cooling systems and networking equipment. Companies must also secure access to increasingly sophisticated semiconductor manufacturing capacity.

Large acquisitions bring additional risks. Reuters reported that Marvell shares fell more than 10% following the announcement, while Lumentum shares dropped more than 15%. The market reaction reflected concerns about the scale of the transaction, its valuation and the financial risks involved.

AMD faces a different challenge: converting growing demand for its AI products into a stronger market position while competing against established players such as NVIDIA. There is also a wider question facing the technology industry: will long-term commercial demand for AI services be strong enough to justify the enormous investments being made in the underlying infrastructure?

What Comes Next for AI Hardware?

AMD's latest results and Marvell's proposed acquisition of Lumentum highlight an important shift in the technology industry. The AI race is moving beyond individual processors and language models towards the infrastructure that allows large computing systems to function.

Powerful CPUs and GPUs remain essential, but so do high-speed networking, optical connectivity, memory and energy-efficient data centres. As AI adoption expands, companies building this underlying infrastructure could become just as important as those developing the models and applications that consumers use.

The next big competition in AI, therefore, may not simply be about who builds the smartest model or the fastest chip. It will also be about who can build the infrastructure capable of processing, connecting and powering the next generation of artificial intelligence.