The AI boom continues to reshape the semiconductor industry, and increasingly, the biggest beneficiaries are not only obvious leaders like Nvidia, but also less visible players at the infrastructure level. One of them is Marvell Technology, which specializes in custom AI chips for the largest cloud providers and expects to turn this area into a multibillion-dollar business in the coming years. The company’s growing role in the AI ecosystem is already being recognized by the market: on June 22, 2026, Marvell is expected to join the S&P 500.
The company said that it expects annual revenue from custom AI silicon for hyperscalers to exceed $10 billion by fiscal 2029, a target that may arrive sooner than investors expect. This is especially significant for Marvell, given that until a few years ago the company remained a niche provider of network solutions and infrastructure components.
Now, AI infrastructure is rapidly becoming the main driver of Marvell’s business. The company has already raised its revenue forecast for the next fiscal year from $15 billion to $16.5 billion and expects to close the current quarter with a result of about $2.7 billion, ahead of analysts’ expectations. The earnings-per-share forecast also came in above market estimates.
Marvell CEO has openly stated that the company works with all U.S. hyperscalers. That is hardly surprising. As the largest cloud providers seek to reduce their reliance on Nvidia’s general-purpose GPUs, they are investing heavily in their own specialized accelerators for specific AI tasks. This trend has significantly increased demand for Marvell’s custom chip design capabilities.
The scale of the potential market is impressive. This year alone, American hyperscalers intend to allocate about $725 billion for the development of computing infrastructure. A significant part of these funds goes not only to the construction of data centers, but also to the development of custom silicon designed to improve energy efficiency, lower inference costs, and optimize the utilization of AI systems.
The financial results already reflect this shift. Marvell’s server business generated $1.83 billion in revenue last quarter, with the segment expanding by 50%. Total quarterly revenue rose 28% to $2.42 billion. In fact, AI infrastructure is becoming the main source of growth for Marvell.
But at the same time, the AI boom is placing increasing pressure on TSMC, whose production facilities remain essential to the AI infrastructure race. The Taiwanese company continues to control more than 70% of the global contract chip manufacturing market, and demand for the 3nm process technology is already beginning to expand beyond the mobile segment.
Previously, smartphone manufacturers represented the primary customers for 3nm lines. Today, Nvidia, Broadcom, Marvell, and suppliers of cloud AI solutions are consuming a growing share of available capacity. According to TrendForce, TSMC’s monthly output of 3nm wafers has already grown to 160,000 to 175,000 units, but the market is still facing a shortage.
Against this backdrop, TSMC is preparing to raise prices again. In the second half of 2026, the cost of manufacturing 3nm chips could rise by as much as 15%, followed by a further increase of 5% to 10% in 2027. For the broader industry, this means higher AI infrastructure costs, as advanced process technologies emerge as one of the sector’s primary bottlenecks.
Additional pressure comes from TSMC’s own investment cycle. The company is simultaneously expanding production in Taiwan, building factories in the USA, Japan, and Europe, and investing in CoWoS packaging, silicon photonics, and multilayer layout technologies. These initiatives require enormous capital expenditures — costs that are gradually being passed on to customers.
TSMC plays a particularly important role for Nvidia and its ecosystem. In addition to the GPUs themselves, the Taiwanese giant provides CoWoS packaging and is developing photonic interconnects needed for the new Spectrum-X and Quantum-X platforms. As AI clusters continue to scale, packaging and interconnections emerge as one of the main factors of productivity.
Taken together, these developments show how quickly AI is turning the semiconductor industry into a new energy infrastructure of the digital economy. Companies like Marvell are becoming suppliers of custom computing solutions for AI data centers, with their shares increasingly appearing among top stock gainers, while TSMC remains a critical manufacturing hub for the entire system.
And while hyperscalers continue to pour hundreds of billions of dollars into computing infrastructure, demand for custom AI chips and advanced technology processes is unlikely to weaken significantly. This means that the shortage of capacity, rising prices, and record profits of chip manufacturers may remain defining features of the industry for years to come.