Broadcom Surges on AI Chip and Networking Boom, Forecasts More Growth

Broadcom Inc. is emerging as one of the biggest winners of the artificial intelligence boom — without selling a single GPU.

Record quarter powered by AI infrastructure

The semiconductor and software company reported record fiscal first-quarter results after U.S. markets closed March 4, driven by surging demand for custom AI chips and the high-speed networks that connect vast data-center clusters. More than $8 billion of its quarterly sales now come directly from AI-related semiconductors, underscoring how a largely behind-the-scenes supplier has become central to the infrastructure powering the latest wave of AI models.

In the quarter ended Feb. 1, Broadcom posted net revenue of $19.31 billion, up 29% from a year earlier. Profit climbed faster: the company reported GAAP net income of $7.35 billion, a 34% increase. Adjusted earnings came in modestly ahead of Wall Street expectations, and adjusted earnings before interest, taxes, depreciation and amortization accounted for 68% of revenue.

The growth engine was largely on the chip side. Revenue from Broadcom’s Semiconductor Solutions segment jumped 52% year over year to $12.52 billion and now accounts for about two-thirds of the company’s business. Infrastructure software, including the VMware unit Broadcom acquired in 2023, was roughly flat at $6.80 billion but continues to generate very high margins.

AI semiconductors now dominate chip sales

AI is now the central theme inside Broadcom’s semiconductor business. The company said revenue tied to AI semiconductors reached $8.4 billion in the first quarter, more than double the level a year earlier and roughly 67% of chip segment sales.

“Broadcom achieved record first quarter revenue on continued strength in AI semiconductor solutions,” President and Chief Executive Hock Tan said in the company’s earnings release. “Q1 AI revenue of $8.4 billion grew 106% year-over-year, above our forecast, driven by robust demand for custom AI accelerators and AI networking. Our AI revenue growth is accelerating, and we expect AI semiconductor revenue to be $10.7 billion in Q2.”

Shares rise as investors weigh a renewed AI thesis

Broadcom shares rose about 4% to 5% in after-hours trading on March 4 following the report and extended those gains the next morning, trading around $330 on March 5, roughly 3% to 5% above the prior close on heavy volume. The move offered a relief rally for a stock that had fallen sharply from highs reached in 2025, even as sentiment toward AI chip names has grown more cautious.

Custom accelerators: broadening the AI chip battlefield

At the heart of Broadcom’s AI surge is a fast-growing business designing custom accelerators — sometimes called XPUs or AI ASICs — for the world’s largest cloud providers and AI labs. These chips are tailored to individual customers’ models and data-center architectures, offering an alternative to off-the-shelf graphics processing units that dominate much of the market today.

On a conference call with analysts, Tan said Broadcom’s custom AI XPU business grew 140% in the quarter compared with a year earlier. While the company does not name all of its customers, it has previously disclosed or strongly hinted at several.

Broadcom designs and supplies chips for Google’s tensor processing units, including the current seventh-generation TPU used in the company’s data centers. Tan said demand for that product remained strong in the quarter. The company is also closely involved in supplying silicon for Anthropic, a startup backed by Amazon and Google, which is planning to ramp AI compute capacity from roughly 1 gigawatt this year to more than 3 gigawatts in 2027.

Meta Platforms, the parent of Facebook and Instagram, is another major customer. Tan used the call to push back on speculation that Meta might scale back its in-house accelerator plans, saying the company’s MTIA chip roadmap is “alive and well” and that Broadcom is already shipping into that program. Two additional hyperscale cloud customers, which Broadcom did not name, are expected to more than double their AI chip volumes in 2027, according to Tan.

In one of the more closely watched disclosures, Broadcom also confirmed that OpenAI has signed on as a custom XPU customer. Tan said the first generation of a Broadcom-built accelerator for OpenAI is expected to be deployed in 2027 at more than 1 gigawatt of compute capacity.

To support those customers, Broadcom has booked substantial manufacturing and packaging capacity at leading chip foundries and memory suppliers. Tan told analysts the company has “secured supply” of advanced wafers, high-bandwidth memory and critical substrates through 2028, giving it visibility into future revenue — and committing it to large volumes of components.

Against that backdrop, Tan said Broadcom now has “line of sight” to more than $100 billion in AI chip revenue in 2027. The company has not publicly broken down whether that figure refers to an annual run rate or a cumulative total over several years, and independent analysts have cautioned that the projection depends on continued aggressive capital spending by a small group of large customers and on favorable market conditions.

AI networking: the fabric connecting the cluster

AI networking is the second major pillar of Broadcom’s AI business. The company sells data-center Ethernet switches, high-speed serializer/deserializer (SerDes) technology and other components that link thousands of accelerators into a single system.

Broadcom said AI networking revenue grew about 60% in the first quarter compared with a year earlier and accounted for roughly one-third of its AI semiconductor revenue, or around $2.8 billion. Tan said that share is expected to rise to roughly 40% in the second quarter, implying about $4.3 billion in AI networking sales if the company meets its $10.7 billion AI semiconductor forecast.

Those sales are anchored by Broadcom’s Tomahawk 6 switch chip, which offers 100 terabits per second of bandwidth, and 200-gigabit-per-second SerDes technology used to drive high-speed optical links. Tan said Broadcom believes it is gaining share in AI networking and is planning to launch a Tomahawk 7 product with about double the performance in 2027. The company is also investing in glass substrates and other packaging advances to maintain signal quality at higher speeds.

Outlook, cash returns and watchpoints

For the current quarter, Broadcom forecast total revenue of about $22 billion, up 47% from the same period a year earlier, and said its adjusted EBITDA margin should remain near 68%. Within that, Broadcom expects semiconductor revenue of roughly $14.8 billion, a 76% increase from the prior-year quarter, with AI semiconductors providing most of the growth.

Kirsten Spears, Broadcom’s chief financial officer and chief accounting officer, told analysts that the company’s mix of high-margin software and growing AI chip revenue supports continued heavy spending on research and development while still returning cash to shareholders.

In the first quarter, Broadcom generated $8.26 billion in cash from operations and about $8.01 billion in free cash flow after capital expenditures. It returned $10.9 billion to shareholders, including $3.1 billion in dividends and $7.8 billion in share repurchases, and its board authorized a new $10 billion buyback program. The company continues to carry a sizable debt load from the VMware acquisition but has emphasized its ability to service that debt with recurring software and maintenance revenue.

Some analysts have responded to the earnings by sharply raising their price targets on Broadcom, citing the breadth of its AI customer base and its sustained profitability. Others have pointed to potential risks, including customer concentration, the cyclical nature of cloud capital spending and the company’s valuation relative to underlying free cash flow.

Broadcom’s regulatory profile also bears watching. Its acquisition of VMware drew scrutiny from competition authorities in the United States and Europe, and enterprise customers have complained about pricing and licensing changes since the deal closed. As Broadcom’s share in AI networking and custom silicon grows, regulators could revisit questions about bundling, interoperability and market power in critical infrastructure.

The scale of the AI build-out Broadcom is helping to enable has implications beyond investors and regulators. Multi-gigawatt deployments of AI compute by companies such as Anthropic, Meta and OpenAI translate into enormous demand for electricity, cooling and physical space. That is already prompting concerns from local governments and grid operators about how to meet rising power needs while staying within climate and environmental limits.

For now, Broadcom is positioning itself as an essential supplier to that expansion. Its latest quarter shows that as AI systems become larger and more complex, value is increasingly accruing not only to the makers of headline-grabbing GPUs, but also to the companies that design the custom chips and dense fabrics that sit underneath.