4 min read

When AI Revenue Eats Your Margins

The Price of AI

There is a simple model for how tech companies are supposed to work: you build a new product, demand spikes, and as you scale up, your margins expand. Operating leverage kicks in, and each new dollar of revenue boosts profits more than the last.

The past week, Broadcom and Oracle revealed a twist in that playbook

Broadcom reported a solid quarter. Revenue beat estimates, and the company raised its guidance, backed by a $73 billion backlog for its AI chips. But the stock fell, partly because of a mechanical detail provided by the CFO. Here is Broadcom's Kirsten Spears explaining why the company's profitability will dip next quarter:

We expect first-quarter consolidated gross margin to be down approximately 100 basis points sequentially, primarily reflecting a higher mix of AI revenue.

This is a specific kind of problem. Broadcom isn't losing money on AI; it is just each dollar of AI revenue is simply less profitable than a dollar from its legacy business. AI products are complex systems, and crucially, they are sold to a tiny group of buyers. Analysts noted that Broadcom's massive backlog comes from just five customers. When your entire growth story depends on five entities who also happen to be the richest, most sophisticated tech companies on earth, your pricing power is naturally constrained.

This concentration risk is becoming a defining feature of the AI hardware market. Oracle's stock drop yesterday—down 11%—was partly driven by anxiety over its reliance on deals with players like OpenAI. If demand from one of these whales changes, the ripple effects are immediate and severe. Even Nvidia, the king of the sector, reported last quarter that over 60% of its revenue came from just four customers (though its monopoly has preserved sky-high margins so far).

And the cost of serving these few, hungry customers is eye-watering. Oracle's capital expenditures jumped to $12 billion for the quarter, pushing its annual forecast to a massive $50 billion. An outlay of that magnitude incinerates cash—Oracle posted a negative free cash flow of $10 billion for the quarter—and it forces a company to tap the debt markets. This has not gone unnoticed by the people who actually lend Oracle the money. Here's Bloomberg:

Wall Street has raised doubts about the costs and time required to develop AI infrastructure at such a massive scale. Oracle has taken out significant sums of debt and committed to leasing multiple data center sites.

The cost of protecting the company's debt against default for five years rose as much as 0.17 percentage point to around 1.41 percentage point a year, the highest intraday level since April 2009, according to ICE Data Services. The gauge rises as investor confidence in the company's credit quality falls. Oracle credit derivatives have become a credit market barometer for AI risk.

This is the new math of the AI infrastructure build-out. For years, tech investors got used to software economics: pure margin, low capital intensity, and infinite scalability. But AI infra is hardware business: capital-intensive, with structurally lower margins and massive upfront spends on land, power, and chips—long before revenue flows.

The market has spent the last year rewarding companies for their exposure to the AI theme. Now it is starting to do the math on the unit economics. Broadcom and Oracle are proving that the growth is real, but investors may have to get comfortable with a different, heavier financial profile to capture it.

More on AI buildout:

  • Brookfield and Qatar launch $20 billion AI infrastructure joint venture (Reuters)
  • SoftBank's Son Eyes Data Center Group Switch to Expand in AI (Bloomberg)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

The Great AI Licensing

  • The Headline: Disney is investing $1 billion in OpenAI and licensing over 200 of its iconic characters for use in the Sora AI video generator, a landmark deal that establishes a new model for how generative AI companies can partner with major IP holders. (WSJ)
  • ARPU's Take: This is a "carrot and stick" strategy from Disney that will reshape the generative AI landscape. By striking a massive deal with OpenAI while simultaneously suing Google for copyright infringement, Disney has just established the new rules of engagement: partner with us and pay, or face our legal wrath.
  • The Product Question: This partnership is a defining moment for generative AI, moving it from a technology that scrapes the web to one that has legitimate, licensed access to the world's most valuable intellectual property. For OpenAI, this deal is a big competitive advantage; it instantly transforms Sora from a generic video tool into a unique, branded content creation engine that rivals cannot replicate without similar, costly deals.

The Gemini Response

  • The Headline: In an accelerated launch reportedly triggered by Google's Gemini 3, OpenAI has released GPT-5.2, a new AI model with improved capabilities in coding, reasoning, and long-context understanding. (Reuters)
  • ARPU's Take: OpenAI is proving that any competitive advantage in the AI model race is measured in weeks, not months. The "code red" and rapid launch of GPT-5.2 is a direct counter-punch to Google's Gemini 3, a clear signal that OpenAI is willing to disrupt its own product roadmap to maintain its perceived technological superiority.
  • The Product Question: This rapid release is a critical move by OpenAI to defend its product leadership and demonstrates an organizational ability to accelerate its R&D roadmap in direct response to competitive pressure. For Google, this potentially neutralizes the short-term performance advantage of its Gemini 3 model.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.


You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.