AI Could Turn Software Into “Dumb Data Pipes” | Dom Rizzo

Dom Rizzo manages the 8.7 billion dollar Global Technology Fund at T. Rowe Price under ticker PRGTX and has returned 43.6 percent per annum since taking over on December 1 2022, outperforming the MSCI All Country World IT index by 5 percent per annum. He attributes that outperformance to being early and in size to AI and semiconductors, guided by a framework focused on linchpin technologies, secular growth markets, improving fundamentals, and reasonable valuation. His central operating thesis is that compute equals revenue, which he uses to explain why the CapEx boom continues and why he estimates the cycle is in inning four or five of a nine-inning game.

Rizzo's most pointed structural claim is that AI will attack enterprise software first because software is a zero incremental cost product whose vendors grew fat on highly recurring, low-churn models that prioritized price increases over user experience. He predicts ChatGPT and Claude will eventually sit on top of the entire enterprise software stack and almost everything else will become a dumb data pipe into those two, with users starting their day with those tools rather than Microsoft Word or Outlook. He estimates 45 billion dollars of recurring run rate revenue is already being crowded out as AI spending competes within fixed IT budgets, and predicts Salesforce's growth rate decelerates from 7 percent rather than accelerating toward 10 percent. Enterprise has not had a true aggregator until now, and he sees OpenAI and Anthropic filling that role.

Agentic computing is distinct from generative AI in that it completes tasks autonomously rather than answering questions, and code generation was the first agentic use case to unlock that capability. A critical hardware implication follows: the GPU to CPU ratio in training workloads is eight to one, but in an agentic world that ratio shifts to parity or two CPUs per one GPU. Because of this, AMD and Intel have been two of the largest bets in his portfolio since the beginning of the year. He estimates the CPU data center market has been roughly 25 billion dollars for some time and expects it to grow to 125 billion dollars.

Rizzo views Nvidia as unequivocally the king of the semiconductor space, framing it as a systems company integrating GPUs, CPUs, and networking rather than a chip company. He calls the Mellanox acquisition the best in semiconductor history because it gave Nvidia best-in-class chip interconnect networking. ARM recently crossed 420 billion dollars in market cap and he sees it as well positioned for agentic AI because its architecture was designed for low-power CPU processing, a key data center requirement. On memory, only Samsung, Hynix, and Micron can make DRAM, and in an agentic AI system memory consumption is roughly five to ten times higher than in prior architectures. Memory sector revenue growth went from roughly zero percent to roughly 500 percent over the past twelve months, though Rizzo describes himself as memory curious rather than fully committed because fundamentals are decelerating and that growth rate must mathematically slow.

Anthropic was doing 5 billion dollars of run rate revenue when Rizzo invested in a funding round last summer. Approximately nine months later Anthropic announced 47 billion dollars of run rate revenue. He credits Dario Amodei and Anthropic for focusing early on chasing scaling laws and understanding code generation. OpenAI and Anthropic combined appear on a path to approximately 200 billion dollars of annualized revenue by year end and are valued at roughly 2 trillion dollars combined, implying approximately ten times revenue. OpenAI's 900 million monthly active users lead Rizzo to predict the company will develop a strong advertising capability over time.

The steel man against the compute equals revenue thesis is that N minus one models are almost as good as leading-edge models but roughly 90 percent cheaper. Rizzo acknowledges every software CEO he has met does hard work at the leading edge and then moves to N minus one as quickly as possible, meaning AI does not behave like the high-margin software model those companies prefer. Scaling laws have held so far, where spending ten times more money yields roughly two times the intelligence, and as long as that relationship holds and demand for leading-edge intelligence remains high, companies are compelled to keep spending.

Semiconductors are up roughly 90 percent year to date while software is flat, and both trade at roughly mid-20s earnings multiples, but semiconductors trade at about nine times sales while software trades at about six times sales, which Rizzo calls historically inverted. Large semiconductor companies trade at PEG ratios of 0.4 to 0.5 times with 30 to 50 percent earnings growth, while the Mag-7 trade at roughly one times PEG. He warns that Apple's services revenue is primarily driven by a single roughly 25 billion dollar annual payment from Google and that if Apple cannot make the shift to agentic operating systems it will be a major problem, with the first agentic operating systems expected in 2027.