Inside the AI engine room: Broadcom’s results and the quiet reshuffle in data-centre chips

Key takeaways

• Broadcom’s 11 December earnings will test whether its rapid artificial intelligence growth can keep pace with lofty expectations.

• The company sits at the crossroads of Nvidia-style GPUs and custom AI chips, giving a clearer view of how the hardware mix evolves.

• Long-term investors can use this print to refine views on AI infrastructure, not to gamble on a single quarter.

Broadcom designs networking chips, custom accelerators and infrastructure software that power many of the world’s largest cloud platforms. It is less visible than Nvidia to consumers, but deeply embedded in how data centres move and process information.

In the third quarter of fiscal 2025, Broadcom delivered strong growth, with revenue up more than 20% and earnings rising over 30%. The company also carries a consolidated backlog of about 110 billion USD, which represents orders already agreed but not yet delivered. That figure will be watched closely this quarter, as it shows whether a small group of giant cloud customers are increasing, holding, or trimming their long-term commitments.

A results day that doubles as an AI check-up

When Broadcom reports fourth quarter fiscal 2025 results after the close on 11 December 2025, investors will really focus on one thing: is the AI infrastructure boom still accelerating, or starting to calm down a little?

Analysts polled by Bloomberg expect quarterly revenue of about 17.5 billion USD, roughly 24% higher than a year ago, and earnings growth of a bit over 30%. Management has guided to around 17.4 billion USD in revenue, in line with that, and has flagged that AI semiconductor revenue could reach roughly 6.2 billion USD. That would be the eleventh straight quarter of AI growth and a strong sign that cloud providers are still spending heavily to build and connect AI servers.

For long-term investors, the appeal is simple. This is not just another earnings call. It is a live snapshot of how the AI chip stack is evolving, from Nvidia’s graphics processors to the quieter rise of custom application-specific chips, and how a “plumbing-heavy” business like Broadcom is trying to secure its place in that value chain.

GPUs, ASICs and the quiet chip reshuffle

Most of the AI story so far has centred on Nvidia’s graphics processing units (GPUs), which are flexible chips originally built for graphics but very well suited to training large AI models. Nvidia still controls roughly 80% to 90% of the AI accelerator market, helped by a strong software ecosystem and tight developer relationships.

However, hyperscalers, the very large cloud providers such as Google and Amazon, are increasingly building their own application specific integrated circuits (ASICs). An ASIC is a chip designed for one main task, such as running recommendation engines or inference workloads. By giving up some flexibility, these chips can improve performance per watt and reduce total cost for very specific jobs.

This is where Broadcom matters. It works with major cloud customers to design custom accelerators and the high-speed networking that links thousands of AI servers together. As Google, Amazon and others move from buying only Nvidia GPUs to a mix of GPUs and in-house ASICs, Broadcom aims to capture a growing share of that custom silicon budget and the cables and switches that keep it all talking.

Industry forecasts suggest that GPUs made around 100 billion USD of data centre revenue in 2024 and that combined GPU and AI ASIC markets could more than double by 2030. The trend is not about GPUs disappearing, but about more specialised chips joining the party. Broadcom’s numbers can offer a clue about how fast that mix is shifting.

What to watch in this week’s numbers

On this earnings call, investors are likely to listen through three main lenses. First, the pace of AI semiconductor growth. If AI revenue lands close to or above the 6.2 billion USD guidance, it suggests the build-out of AI data centres remains robust. A materially softer number would support the idea that the first phase of the AI investment surge is easing.

Second, the mix of demand. Management commentary on which customers are growing fastest, whether spend is focused on training or inference, and how much is going to custom chips versus off-the-shelf parts, will help frame the balance between Nvidia-style platforms and ASIC-heavy strategies.

Third, discipline. Investors will compare AI growth with updates on capital expenditure, margins and cash returns. Broadcom has presented itself as a relatively steady compounder inside a very volatile AI story. The more it can show growth with stable or improving profitability, the stronger that narrative becomes for long-term holders.

Risks to keep in view

There are real risks around this optimistic AI infrastructure picture. The most obvious is a pause in spending. If cloud providers decide they have built enough capacity for now, or if end-user AI adoption disappoints, orders for both GPUs and custom chips could slow. That would hit Broadcom’s growth at the same time as its customers digest past investment.

Competition is another risk. Nvidia is not standing still, and rivals such as AMD and others are pushing their own platforms. At the same time, hyperscalers may use their purchasing power to pressure suppliers on pricing. Broadcom also remains exposed to integration and execution risk around VMware, and to customer concentration, given how much of its backlog depends on a handful of very large buyers.

Investor playbook

• Treat Broadcom as one window into the AI infrastructure theme, not a complete picture of AI investing.

• Listen for clues on the balance between GPUs and custom ASICs when reviewing the earnings commentary.

• Focus on the health of the backlog, margins and cash generation rather than the exact earnings-per-share figure.

• Use scenarios and position sizing, so any single AI supplier does not dominate your portfolio risk.

One print, a much bigger story

Broadcom’s earnings this week are about far more than whether it beats consensus by a small amount. They are a real time test of how the AI hardware stack is evolving, how confident the biggest cloud players feel about future demand, and whether there is room for both Nvidia-style GPU platforms and quieter custom chips to thrive together.

For long-term investors, the most useful takeaway may be perspective. The AI infrastructure cycle is likely to move in waves, and different players will lead at different times. Watching Broadcom’s results alongside those of Nvidia, AMD and others can help you track where value is shifting in the stack. The goal is not to chase whichever acronym is hottest, but to build a portfolio that can live with the twists of a long AI build-out.

This material is marketing content and should not be regarded as investment advice. Trading financial instruments carries risks and historic performance is not a guarantee of future results.

The instrument(s) referenced in this content may be issued by a partner, from whom Saxo receives promotional fees, payment or retrocessions. While Saxo may receive compensation from these partnerships, all content is created with the aim of providing clients with valuable information and options.