In the long arc of technological transformation, few moments are as telling as when the infrastructure layer becomes indispensable — and Broadcom finds itself precisely there, supplying the custom silicon that the largest AI companies are quietly building their futures upon. Despite losing a quarter of its market value since June 2026, the semiconductor giant's AI revenue surged 221 percent year-over-year, and its chief executive projects that figure will reach $115 billion by fiscal 2027 and double again by 2028. The company's story is less about a stock price recovering and more about whethe
Broadcom positioned to rejoin $2 trillion club by 2027 on AI chip momentum
Demand is so strong they can't make chips fast enough
So Broadcom was already in the $2 trillion club and fell out. What changed?
The stock dropped 23 percent from its June peak because investors got nervous about whether companies would keep spending on AI infrastructure at the same pace. Some big tech companies started rationing AI use and switching to cheaper models.
But that's just caution, not actual evidence of a slowdown in Broadcom's business, right?
Right. Their latest earnings show AI revenue jumped 221 percent year-over-year. The company says demand is so strong they're actually supply-constrained—they can't make chips fast enough.
So the custom chips—the TPUs and Jalapeño—those are the real growth engine?
Exactly. Instead of buying Nvidia's generic GPUs, companies like Alphabet and OpenAI are designing accelerators optimized for their specific models. Broadcom manufactures them. It's a different business model.
How much of Broadcom's revenue is actually coming from these custom chips versus traditional data center components?
In the third quarter, AI semiconductors were $16.7 billion of their $29.6 billion total revenue. Management is forecasting $115 billion in AI revenue for fiscal 2027.
And that's why they think they'll hit $2 trillion?
The math works if those forecasts hold. At $173 billion in projected fiscal 2027 revenue, the stock trades at a forward price-to-sales of 10.1, which is reasonable historically.
But that's all contingent on Broadcom actually delivering those numbers and the market staying convinced about the custom accelerator thesis. If either breaks, the stock could fall further.
True. The company is betting everything on supply and demand staying in balance, and on these custom chips becoming the standard rather than a niche play.
Der Puls
- Broadcom's stock has shed 25% of its value since June, falling from a $2 trillion valuation to $1.77 trillion amid investor fears that the AI infrastructure boom may be losing steam.
- Beneath the surface anxiety, a structural shift is accelerating — Alphabet, OpenAI, and Anthropic are abandoning generic Nvidia GPUs in favor of custom accelerators that Broadcom is designing and manufacturing for them.
- AI semiconductor revenue hit $16.7 billion in Q3 — a 221% year-over-year surge — with management projecting $21.7 billion in Q4, signaling that demand is not slowing but that manufacturing capacity is the only real constraint.
- CEO Hock Tan has told investors that AI revenue could reach $115 billion in fiscal 2027 and $230 billion in 2028, framing the company's ceiling not as a demand problem but a supply problem.
- The valuation case for a return to $2 trillion rests on a forward price-to-sales ratio of 10.1x — right at Broadcom's historical average — if fiscal 2027 revenue projections of $173 billion hold.
In the long arc of technological transformation, few moments are as telling as when the infrastructure layer becomes indispensable — and Broadcom finds itself precisely there, supplying the custom silicon that the largest AI companies are quietly building their futures upon. Despite losing a quarter of its market value since June 2026, the semiconductor giant's AI revenue surged 221 percent year-over-year, and its chief executive projects that figure will reach $115 billion by fiscal 2027 and double again by 2028. The company's story is less about a stock price recovering and more about whether the age of generic computing hardware is giving way to something more specialized, more sovereign, and more permanent.
Broadcom has spent the past year becoming the quiet backbone of the artificial intelligence industry, manufacturing the data center chips that keep large language models running. Yet despite that central role, its stock has lost roughly a quarter of its value since June 2026, dragged down by investor anxiety about whether the AI infrastructure boom can sustain itself. Once worth $2 trillion, the company now sits at $1.77 trillion — and the question is whether it can climb back.
The skepticism has surface logic: major tech companies have begun rationing AI usage and shifting toward cheaper models to control costs. But Broadcom's most recent earnings suggest a more important story is unfolding beneath the headlines. The largest AI companies are moving away from generic chips and toward building their own — and Broadcom is the partner helping them do it.
Alphabet has a multi-generation partnership with Broadcom to manufacture its Tensor Processing Units, which power Gemini, are rented through Google Cloud, and are now sold directly to customers like Anthropic. Broadcom recently began shipping Alphabet's v8i TPU, which the company claims outperforms Nvidia's latest Vera Rubin GPU. OpenAI took a separate path, working with Broadcom to design a custom accelerator called Jalapeño, built specifically for its workloads. The strategic logic is significant: chips optimized for a particular model outperform general-purpose GPUs, and they break the pricing leverage that Nvidia has long held over the industry.
The revenue numbers reflect how fast this is moving. In Q3, Broadcom posted $29.6 billion in total revenue — an 86% jump year-over-year — with AI semiconductor sales alone hitting $16.7 billion, a 221% increase. Net income more than tripled to $13.1 billion. CEO Hock Tan told investors that AI revenue could reach $115 billion in fiscal 2027 and double to $230 billion in fiscal 2028, with the only constraint being manufacturing capacity, not demand.
The path back to a $2 trillion valuation runs through that forecast. On a forward basis, using projected fiscal 2027 revenue of $173 billion, Broadcom's price-to-sales ratio falls to 10.1 — right at its historical average. Whether the market rerate happens depends on whether Broadcom executes on Tan's targets and whether confidence in the custom accelerator strategy continues to build.
Broadcom has spent the last year becoming indispensable to the artificial intelligence industry. The semiconductor company manufactures the data center chips and components that power AI training and inference—the computational backbone that keeps the large language models running. Yet despite this central position, Broadcom's stock has lost a quarter of its value since June, dragged down by investor anxiety about whether the AI infrastructure boom can actually sustain itself. The company was once worth $2 trillion. Today, as of early September 2026, it sits at $1.77 trillion. The question now is whether it can climb back.
The skepticism makes surface sense. Over recent months, major technology companies have begun rationing AI usage among their employees and shifting workloads toward cheaper models to control costs. When spending discipline takes hold in the industry, semiconductor demand typically follows. But Broadcom's most recent quarterly earnings suggest the company still has substantial runway ahead. The real story isn't in the headlines about caution—it's in the shift happening beneath them, where the largest AI companies are moving away from buying generic chips and toward building their own.
Nvidia's graphics processing units remain the default choice for most AI work. They are powerful, proven, and widely available. But Alphabet, OpenAI, Anthropic, and Meta Platforms have all begun designing custom accelerators tailored to their specific needs, and Broadcom is the partner helping them build these specialized chips. Alphabet has a multi-generation partnership with Broadcom to manufacture its Tensor Processing Units, or TPUs. The company uses these chips to improve its Gemini AI models, rents them to developers through Google Cloud, and now sells them directly to customers like Anthropic. Broadcom recently began shipping Alphabet's new v8i TPU, which the company claims outperforms Nvidia's latest Vera Rubin GPU. The momentum is expected to accelerate through 2028 and 2029, with Anthropic positioned to become Broadcom's largest customer next year.
OpenAI took a different path. Rather than licensing Alphabet's TPUs, it worked with Broadcom to design an entirely separate accelerator called Jalapeño, built specifically for OpenAI's workloads and comparable in performance to Vera Rubin. This matters more than it might initially appear. When an accelerator is optimized for a particular large language model, Broadcom argues, it outperforms any general-purpose GPU. More importantly, it breaks the lock-in that companies have historically faced with suppliers like Nvidia, where pricing and innovation remain controlled by a single vendor. Custom chips give AI companies leverage they didn't have before.
The revenue numbers tell the story of how fast this is moving. In Broadcom's third quarter, which ended in early August, the company generated $29.6 billion in total revenue—an 86 percent jump from the same quarter a year earlier. That growth rate itself accelerated sharply from the previous quarter, when revenue had climbed 48 percent. The driver was unmistakable: AI semiconductor sales hit $16.7 billion, a 221 percent increase year-over-year. Management forecasted that AI semiconductor revenue could grow even faster in the current fourth quarter, potentially reaching $21.7 billion and representing 236 percent growth. The company's net income more than tripled to $13.1 billion in the quarter.
Broadcom's chief executive, Hock Tan, told investors that AI revenue could reach a record $115 billion in fiscal 2027, then double to $230 billion in fiscal 2028. He attributed the ceiling on growth not to demand—which he described as overwhelming—but to manufacturing capacity. If Broadcom could produce more chips, he said, it would immediately sell them. The company is essentially supply-constrained in an environment where customers are desperate to buy.
The valuation math is where the $2 trillion prediction takes shape. On a trailing-12-month basis, Broadcom's stock trades at 20.2 times sales, nearly double its 10-year average of 10.5. By that measure, the stock looks expensive. But Wall Street analysts project the company will generate $173 billion in total revenue during fiscal 2027. Using that forward projection, the price-to-sales ratio drops to 10.1—back to the historical average. If Broadcom executes on Tan's forecast to double AI revenue in 2028, the valuation becomes even more attractive. The path to $2 trillion depends on whether those growth targets hold and whether the market's confidence in the custom accelerator strategy continues to build.
Bemerkenswerte Zitate
A lack of supply is the only reason growth won't be even faster—if the company finds a way to manufacture more chips, it could immediately generate additional revenue because demand is so strong.— Hock Tan, Broadcom CEO