Broadcom Tripled AI Chip Sales and the Stock Still Fell. Is AVGO the Nvidia Alternative Investors Missed?

Broadcom just posted AI chip growth that dwarfs nearly every semiconductor result this year, yet the stock sold off anyway. The real question is whether the market priced in the wrong number entirely.

Published September 3, 2026, 1:22pm ET · 3 min read

A close-up, high-angle view of a dark grey semiconductor chip with the white letters 'AI' (Artificial Intelligence) printed on its surface, centrally positioned on a dark blue circuit board. The circuit board is densely populated with various smaller electronic components, including resistors and capacitors, and features intricate copper traces. The entire assembly is illuminated with a gradient of cool blue light on the left transitioning to warm pink-purple light on the right.
A prominent 'AI' chip on a circuit board signifies the crucial role of artificial intelligence in driving the semiconductor industry, a key area for Broadcom's market position. © Quality Stock Arts / Shutterstock.com

Broadcom (NASDAQ:AVGO | AVGO Price Prediction) reported a quarter that, on the numbers, looks like one of the most consequential AI results of the year. AI semiconductor revenue reached $16.7 billion, up 221% year-on-year and 54% sequentially, and management raised the fiscal 2027 AI revenue outlook to roughly $115 billion, with fiscal 2028 pointing to $230 billion.

The post-earnings reaction was still negative, largely because the Q4 total revenue guide of about $34.8 billion landed close to, but not clearly above, consensus. The settled session closed down 0.66% at $367.24, with the sharper move happening in extended hours around the report.

AVGO earnings explorer

What Broadcom Actually Reported

Revenue came in at $29.591 billion, up 85.5% year-over-year, with non-GAAP EPS of $3.32 against a $3.2382 estimate. Operating margin expanded 240 basis points to 67.9%, and free cash flow was $13.665 billion, or 46% of revenue.

The market chose to trade the one line that disappointed: the Q4 top-line guide. That is the reasonable-sounding overreaction, because the disclosure that mattered more was the forward AI trajectory.

A custom AI accelerator, or XPU, is silicon co-designed with a specific hyperscaler for a specific model family. XPUs represented 73% of AI revenue and XPU shipments grew more than three and a half times year-on-year. Hock Tan put it directly: “When you co-develop a chip that is optimized for your particular LLM workloads, you will outperform any GPU.”

NVIDIA (NASDAQ:NVDA) sells a general-purpose platform with the CUDA ecosystem attached. Broadcom plays a different role, supplying design expertise and networking for hyperscalers who already know what they want to run.

Gigawatt Visibility and Customer Concentration

Management named the deployments. Anthropic is deploying one gigawatt of Ironwood in 2026, another five gigawatts of TPU v8i in 2027, and a further 10 gigawatts in 2028. OpenAI has Jalapeno on track for 1.3 gigawatts in 2027 and over five gigawatts across generations in 2028. Gigawatts indicate ordered capacity that still depends on power, land, and packaging before it converts to revenue. The concentration is real: six XPU customers, with four expected to be particularly large. Any one of them changing plans matters.

The underappreciated piece is networking. Tomahawk 6 is a first-to-market 100-terabit switch deployed by pretty much all AI hyperscalers building XPUs with Broadcom, and Tomahawk 7 is taped out as the industry’s first 200-terabit-per-second Ethernet switch. The mix shift toward custom silicon dilutes blended gross margin because memory content is heavier, which is why Q4 gross margin is guided to about 73%. That is a feature. A hyperscaler that co-designed a chip cannot move that program without restarting a multi-year design cycle.

Is Broadcom Stock a Buy?

AVGO price target

AVGO analyst ratings

NVIDIA still owns general-purpose accelerated computing and the developer ecosystem, trading at a 45x P/E. Marvell (NASDAQ:MRVL) is the other merchant custom-silicon option but at a smaller scale. Broadcom’s forward P/E near 20x, against a $525.97 analyst target, and Tan’s stated goal of exceeding $30 in EPS in fiscal 2028 make the risk-reward profile attractive despite concentration.

The framing of an Nvidia alternative is the whole point, and we cataloged the early traits the biggest tech winners shared in a free playbook here: The Next Nvidia Playbook. The setup screens as attractive on a research basis, though the customer-concentration discount argues for measured position sizing.

Contact [email protected] for any questions or corrections.

Omor Ibne Ehsan

Omor Ibne Ehsan is a writer at 24/7 Wall St. He is a self-taught investor with a focus on growth and cyclical stocks that have strong fundamentals, value, and long-term potential. He also has an interest in high-risk, high-reward investments such as cryptocurrencies and penny stocks.

All articles →