On a Monday afternoon, the filing—the kind of paperwork that most people ignore—arrived. However, by Tuesday morning, Broadcom’s stock had increased roughly 3.5% in premarket trading, passing $325.28, and traders were subtly changing their perceptions of the company’s true nature. The fact that a single SEC document—which is dense and unglamorous—can benefit a stock more than a spectacular product launch is telling. Investors had been anticipating evidence. They understood.
The filing’s confirmation wasn’t totally novel. Since about 2016, when Google was still the only significant corporation bold enough to develop its own AI chips internally, Broadcom and Google have been collaborating on the Tensor Processing Unit. That wager, placed almost ten years ago, now appears remarkably prophetic. The new deal brings Broadcom’s networking equipment closer to Google’s data centers and extends the partnership through 2031. Analysts are reminded of the Apple-Broadcom agreement, which has quietly produced billions for years, by this kind of long-term, sticky commitment.
| Field | Detail |
|---|---|
| Company | Broadcom Inc. |
| Ticker | NASDAQ: AVGO |
| CEO | Hock Tan |
| Headquarters | Palo Alto, California |
| Market Capitalization | Briefly crossed $2 trillion in April 2026 |
| Key Partner | Alphabet’s Google, top customer since 2016 |
| Latest Product Tie-In | Google’s 8th-generation TPU (TPU 8t and TPU 8i) |
| Filing Reference | SEC Form 8-K filing, April 2026 |
| Stock Movement | Premarket rise of about 3.5% to $325.28 after filing |
| Year-to-Date Performance | Down roughly 6% earlier in 2026, recovered after announcement |
Harlan Sur of JPMorgan referred to it as a “major proof point,” and it makes sense. The possibility that Google would eventually attempt to bring more of this work in-house, eliminating Broadcom, had caused the market to become uneasy. That concern seems exaggerated, at least for the time being. It appears that Google has determined that the collaboration is worthwhile, possibly because creating cutting-edge chips at Broadcom’s size is more difficult than it first appears.
The story becomes more intriguing during the eighth-generation TPU rollout. Unusually, Google divided it into two chips this time. Training, the labor-intensive process by which models learn, is handled by TPU 8t. TPU 8i is designed for inference, which is the lighter, quicker task of responding to queries and assisting users. Back in March, Hock Tan alluded to this change by stating that clients would begin creating two chips annually, one for each job. Almost exactly as he described, the roadmap came to pass. When a CEO makes a public prediction and then sees it come to pass a month later, it is difficult to ignore.
The split is a quiet windfall for Broadcom. The aspect of AI that grows with regular use, searches, chatbot responses, and recommendations is called inference. According to Google, the TPU 8i outperforms the previous generation by 80% in terms of performance per dollar, which translates into less expensive inference, which typically translates into more inference. More chips, more volume. The math is in Broadcom’s favor.
Investors were surprised to learn that Anthropic, the AI startup responsible for Claude, had committed to several gigawatts of next-generation TPU capacity through Google and Broadcom. That’s a big deal. This kind of commitment ties Broadcom into the supply chains of yet another significant model developer and indicates where the AI industry’s compute demand is actually headed next year.
Earlier this year, Broadcom’s stock fell roughly 6% as a result of the same tense trade that caused Nvidia and AMD to decline. Investors’ doubts about the sustainability of the spending boom have caused the chip industry to falter. On April 22, Broadcom’s market capitalization momentarily reached $2 trillion before slightly declining. It’s still unclear if the rally will continue. For the time being, however, the filing accomplished what filings seldom do. People were forced to pay attention.