A hyperscale data center has a unique sound that is uncommon in many other industrial settings. It’s not precisely the steady hum of transformers or the high-pitched whine of cooling fans. The gentle click of optical interconnects firing every few microseconds is layered on top of both. The air feels artificial when you walk through one in central Iowa or northern Virginia. The temperature is deliberately incorrect. Humidity remains artificial. The reason why a few stocks had such an amazing April can be found inside those buildings.
Up until April 2026, five names dominated the AI discourse. AMD, Nebius, CoreWeave, Broadcom, and Nvidia. distinct products, distinct histories, and distinct investor pitches. However, their stock charts moved in a formation-like manner last month. The explanation is almost embarrassingly straightforward. The silicon and capacity that hyperscalers are unable to stop purchasing are all being sold, leased, or designed by them. That’s all. For now, that’s the whole story.
The weird center of the universe is still Nvidia. The company’s revenue increased by 73% in its fiscal 2026 fourth quarter, which ended on January 25. The management is aiming for 77% in the following quarter. That degree of acceleration is so out of character for a company with a $4.8 trillion market capitalization. Jensen Huang’s repeated remarks regarding $500 billion in advanced chip bookings through 2026 support investors’ apparent belief that the buildout is genuine. At any conference on AI infrastructure, there’s a feeling that no one really knows how long this will last, but no one wants to wager against it either.
Despite playing a different game, Broadcom manages to win it just as easily. Instead of using general-purpose GPUs to take on Nvidia head-on, Hock Tan’s team has quietly built an empire around custom AI ASICs that were created in close coordination with hyperscale clients. Although Broadcom’s overall growth was about 29% in the most recent quarter, investors are now primarily focused on the segment of custom AI silicon. By the end of 2027, management hopes to generate $100 billion in revenue from custom chips annually. For fiscal 2026, Wall Street analysts predict 64% overall growth, which will only marginally slow to 50% the following year. This transformation into an AI silicon partnership is one of the more underappreciated rewrites in recent tech history for a company that began by selling fiber-optic transceivers.
The wild cards, Nebius and CoreWeave, reveal what’s actually going on beneath the surface. Both are neoclouds, an awkward term used by the industry to describe cloud service providers that are primarily focused on renting GPUs. Nebius, which has a special supply agreement with Nvidia, anticipates that by the end of 2026, its annualized run rate will have increased from $1.25 billion at the end of 2025 to between $7 billion and $9 billion. CoreWeave, which provides GPU compute to customers like Microsoft and Meta, is expected to increase revenue by 142% this year. Neither name appeared on the majority of institutional buy lists three years ago. They are now considered vital infrastructure in discussions with Amazon and Microsoft.
Since AMD is no longer an option, it should be on the list. The MI325 gained popularity, the MI450 has started shipping into customer environments, and the MI300 line opened the door. Lisa Su’s company has led its data center division to more than 60% compound annual growth. It seems like AMD has finally emerged from the shadow that surrounded it for twenty years. For the time being, the trajectory is what counts, but it’s still unclear if margins can withstand Nvidia’s pricing power in the long run.
Annualized AI data center revenue growth above 50% is the common metric that sits beneath everything else. When you remove the various business models, patent portfolios, and marketing jargon, all that’s left is a single figure that is growing at a remarkable rate. It’s difficult to ignore how highly concentrated the AI market is. Five businesses. Just one metric. One source of demand, primarily financed by a few trillion-dollar purchasers who consistently increase their capital expenditure goals. History will determine whether this appears to be a true industrial transformation or an eventual overbuild. The response was straightforward enough to be contained in a single line item, at least for April.
