Broadcom rallies 40% as custom AI networking silicon and Tomahawk switches gain hyperscaler adoption, outpacing broader chip sector.

Broadcom has quietly overtaken Nvidia as the hotter semiconductor trade over the past year, powered by a surge in demand for custom artificial intelligence chips that threatens to loosen Nvidia's grip on the AI infrastructure market. While Nvidia remains the undisputed king of general-purpose AI accelerators, Broadcom's stock has returned roughly 40% over the past 12 months, more than doubling Nvidia's 22% gain, and has climbed 36% in just the last six months.

The divergence marks a pivotal moment in the AI chip wars. Nvidia's graphics processing units are the standardized workhorses that train and run massive AI models inside data centers around the world. Broadcom, by contrast, designs application-specific integrated circuits, or ASICs, that are custom-built for a single hyperscaler's most critical workloads. As Alphabet, Meta Platforms, Microsoft and Amazon double down on in-house silicon to reduce their dependence on Nvidia's pricey platforms, Broadcom's order book is swelling at an extraordinary pace.

In its fiscal 2026 second quarter, Broadcom reported total revenue of $22.19 billion, up 48% from a year earlier, while net income nearly doubled. The standout figure was AI semiconductor revenue, which jumped 143% year over year to $10.80 billion and now accounts for almost half of total sales. CEO Hock Tan told investors that AI semiconductor revenue is expected to grow more than 200% in the fiscal third quarter, reaching $16 billion. That single segment would then represent the majority of the company's projected $29 billion in overall quarterly revenue.

Nvidia is hardly standing still. The company posted first-quarter fiscal 2027 revenue of $81.62 billion, an 85% increase from the prior year. Data center sales alone hit $75.25 billion, and networking revenue inside that unit surged 199%, underscoring how Nvidia's CUDA software ecosystem has evolved from selling chips into selling entire rack-scale AI factory systems built on InfiniBand, NVLink and Spectrum-X. Guidance for the second quarter calls for $91 billion in revenue, excluding any contribution from China, with a non-GAAP gross margin of 75%.

Yet the growth rates tell a nuanced story. Nvidia's data center revenue rose 92% year over year in the same period, a blistering pace by any standard. Broadcom's 143% AI silicon growth, however, suggests it is capturing market share at a faster clip inside the most coveted slice of the AI chip industry. Analysts have taken note: Nvidia carries a consensus price target of roughly $303, implying about 38% upside from recent levels, with 58 of 61 analysts rating the stock a Buy or Strong Buy. Broadcom's target near $528 implies roughly 25% upside, though its recent outperformance has compressed that gap.

The valuation picture further complicates the comparison. Nvidia trades at a forward price-to-earnings multiple of 23 with a price-to-earnings-to-growth ratio of 0.55, a level that looks downright cheap next to Broadcom's forward P/E of 69. Nvidia generated $48.55 billion in free cash flow in a single quarter and $96.58 billion for the full fiscal year, and it carries an $80 billion share buyback authorization alongside $119 billion in supply commitments that signal management expects the AI capex cycle to extend past 2027.

Broadcom's bulls argue that the P/E premium is justified because the company is riding a structural shift. Hyperscalers are no longer content to buy every GPU Nvidia can manufacture. Alphabet's Tensor Processing Units, or TPUs, and Meta's MTIA chips are both designed with Broadcom, and Alphabet recently disclosed that it has begun selling TPUs to select outside customers. That move, announced during Alphabet's second-quarter earnings call, puts the search giant in direct competition with Nvidia while simultaneously enlarging Broadcom's addressable market. Alphabet and Broadcom have a development deal that runs through 2031.

Alphabet CEO Sundar Pichai said earlier this year when he first flagged the plan to commercialize TPUs: "Custom chips are cheaper than Nvidia's GPUs and highly effective at handling the workloads for which they were designed." Google Cloud revenue hit $24.8 billion in the most recent quarter, an 82% jump, and the unit ended the period with a $514 billion backlog, providing a deep well of demand for whatever silicon Alphabet chooses to deploy.

The ASIC-versus-GPU debate is not winner-take-all. Data centers will need both for the foreseeable future. An ASIC can perform a single task with greater energy efficiency and lower latency than a GPU, which matters enormously for AI inference, the process of running a trained model to deliver answers. A GPU, however, can handle a far wider range of tasks, making it indispensable for the training phase and for cloud providers that serve a diverse set of customers. Nvidia's CUDA platform, with its vast library of optimized software, remains the default choice for most AI developers.

Through the risk-adjusted lens of Bridgewater Associates founder Ray Dalio, Nvidia at 23 times forward earnings may still be the cleaner bet. Dalio's philosophy favors durable cash flow, reasonable pricing and setups where growth is real and defensible. Nvidia's 114% return on equity and 75% gross margins fit that framework. Broadcom, by contrast, carries customer concentration risk because a handful of hyperscalers account for a large share of its custom-chip revenue, and its VMware acquisition added substantial debt to the balance sheet.

Micron Technology, the third leg of the AI chip stool, is an even more extreme case. The memory maker's stock has skyrocketed more than 700% over the past year as high-bandwidth memory became a must-have component for Nvidia's GPUs. Micron trades at a forward P/E of just 5 with a PEG of 0.12, numbers that scream bargain. But memory is notoriously cyclical, and the 711% one-year gain already prices in a supply-tight cycle that can reverse with little warning.

Broadcom's software division, anchored by VMware, provides a $7.18 billion quarterly revenue cushion that neither Nvidia nor Micron possesses. That diversification helps insulate Broadcom if AI chip spending ever cools. The company's adjusted EBITDA margin of 69% rivals Nvidia's 75% gross margin, suggesting both firms extract enormous profitability from the AI buildout.

The next catalyst for both stocks arrives on August 26, when Nvidia reports fiscal second-quarter results. The $91 billion revenue bar is the number to watch, alongside whether data center networking growth stays above 100%. For Broadcom, the third-quarter AI revenue guide of $16 billion is the tell. A miss would puncture the custom-silicon narrative; a beat would lend credence to the idea that Broadcom is evolving from challenger to co-champion.

Wall Street's positioning reflects the tension. Polymarket traders recently assigned only a 57% probability that Nvidia shares would close a week above $220, and a seven-day sentiment score for the stock fell nearly seven points, signaling fading conviction ahead of earnings. Broadcom's composite sentiment score sits at a bearish 36.19, suggesting the market is already pricing in a high bar for execution.

For investors, the choice between the two chip giants hinges on which thesis they believe. Nvidia's bet is that AI workloads will continue to favor general-purpose accelerators wrapped in a software ecosystem that competitors cannot easily replicate. Broadcom's bet is that the world's largest technology companies will keep pouring capital into custom designs to escape Nvidia's pricing power. Both narratives can coexist, and the outcome will likely depend on how rapidly hyperscalers can develop and deploy custom silicon at scale.