Broadcom is a buy because of its growth from AI and its partnership with technology giant Google (NASDAQ: GOOG).
San Jose, California-based Broadcom (NASDAQ: AVGO), a designer, developer, manufacturer and supplier of infrastructure software and semiconductor products, is worth considering as a buy due to capitalizing on the advent of artificial intelligence (AI), cutting-edge collaboration with Google and its other growth opportunities. Investors looking for a technology stock that offers income also may appreciate Broadcom’s modest dividend yield of 0.78% and a dividend that has been increased for 15 consecutive years.
Major customers of Broadcom include Anthropic, totaling $21 billion in recent semiconductor orders, and OpenAI, a buyer of $10 billion in Broadcom semiconductors. Broadcom designs AI chips to meet the requirements of both Anthropic and OpenAI. Semiconductors, also called computer chips, are critically important for technological companies and users of AI.
Broadcom Is a Buy Because of Its Growth from AI: Customized Computer Chips
Broadcom’s business model is to collaborate with AI-focused companies to design customized computer chips, which are manufactured by foundries such as Taiwan Semiconductor Manufacturing Company (NYSE: TSM). Consider Broadcom’s role as a provider of intellectual property that is used in computer chips to help its customers benefit from AI.
Customized AI Chips Are Lifting Google
Customized chips designed to use AI are aiding Google. One way is that AI is attracting huge amounts of investment funds intended to tap the power of this breakthrough technology. Google, a part of Alphabet Inc. (NASDAQ: GOOG) is one of the early beneficiaries of the new capabilities offered by AI.
One of the key drivers to Broadcom’s new sales to Anthropic and OpenAI during its last quarter was a broader shift in how the AI industry evaluates hardware choices. And that shift became more visible after Google’s Gemini 3 reportedly demonstrated competitive outperformance against OpenAI’s GPT-5.1 in key benchmarks.
Not only did Google outperform GPT with the new AI chip, but Broadcom benefited by winning orders from other tech companies, too.
The Blue Chip Stock Trader advisory service, led by Jim Woods, has helped his subscribers profit by recommending Alphabet. He advised buying both its stock and related call options on November 20, 2025. Each has climbed quickly by double-digit percentages.
Chart courtesy of www.stockcharts.com
Gemini 3 was trained on Google’s in-house TPUs (Tensor Processing Units), the custom AI chips developed in collaboration with Broadcom, rather than on GPUs (Graphic Processing Units) supplied by NVIDIA. That result helped validate Custom AI Chips as a viable alternative for large-scale AI training, allowing big AI platform companies to plan Capex more deliberately instead of endlessly adding default GPUs.
Broadcom Is a Buy Because of Its Growth from AI: Bet on Compute
Companies and investors all tend to agree that more compute — the scale of data center resources committed to AI training — lead to better AI. But why has this belief become so widely accepted? The confidence traces back to a powerful idea introduced by OpenAI: Scaling Law.
The concept of the Scaling Law originates from a 2020 study by the OpenAI research team, demonstrating that bigger model, larger dataset and more compute led to predictable and consistent performance improvements.
The chart is based on the data from Scaling Laws for Neural Language Models and Epoch AI
This dynamic is illustrated in the figure above, which estimates the additional compute required to achieve a further 1% reduction in training loss across successive generations of models, from GPT-3 through GPT-5. Extra compute needed means more GPU counts — more orders for NVIDIA.
Of course, GPUs were flexible and ideal for AI model training. But as models scaled, the question changed. It was no longer whether more compute was needed, but whether GPUs were the most economically efficient way to deliver it. That shift has pushed custom AI chips, known as ASICs (Application-Specific Integrated Circuits), into the spotlight. At the same level of compute power, ASICs consume fewer resources.
Source: Generated by Gemini Nano Banana
From GPU Spending to Disciplined Capex
The advantage of custom AI chips extends beyond performance and efficiency; it further fundamentally changes how AI infrastructure is financed. By narrowing hardware functionality to well-defined workloads, ASICs make compute spending more predictable and easier to plan over time.
Source: Quarterly financial reports from Meta, Google and Microsoft
From a financial perspective, capital expenditures (Capex) among leading AI companies have risen steadily in recent years. These investments primarily reflect data center construction, compute hardware and supporting infrastructure such as power and cooling.
When Capex is spent on NVIDIA GPUs, companies often depreciate NVIDIA GPUs over 5–6 years for accounting purposes, even though the effective economic life of these chips in heavy AI workloads may be closer to 2–3 years. Also, with pricing set by NVIDIA and flagship chips now costing tens of thousands of dollars each, GPU-based Capex is very difficult to control. Directing Capex toward ASICs changes that trade-off. While ASICs require more initial investment, they reduce long-term reliance on NVIDIA’s pricing.
The Hidden Challenges of Custom AI Chips
The first constraint is manufacturing capacity. Shifting to ASICs means re-engaging the entire supply chain. Beyond chip design, ASICs depend on power availability, advanced foundry capacity at TSMC, high-bandwidth memory (HBM), advanced packaging and long-term infrastructure investment. The key question is not whether ASICs can lower unit costs in theory, but whether enough suppliers are willing to commit Capex to support new architectures at scale. After all, NVIDIA’s president and chief executive officer Jensen Huang frequently flies to Asia for business trips.
And there is also the question of partnership dynamics. Broadcom’s partnership with Google does not necessarily deliver high margins for Google, as Broadcom captures much of the system-level economics. Google has explored other hardware partnerships with MediaTek Inc., but replacing Broadcom in its custom AI chip program would introduce uncertainty around technology control and the pace of execution. At this stage, such partnerships function more as a backup than a true substitute.
Looking beyond the headlines, Broadcom’s fundamentals remain solid. In FY2025, the company generated approximately $27.5 billion in cash flow, while capital expenditures were just $6.2 billion, providing flexibility to invest in Research and Development, while continuing to return capital to shareholders. At the same time, its order pipeline over the next several quarters appears well supported. It will be worth watching how Broadcom — or custom AI chips more broadly — perform in the period ahead.
