Gabelli Funds portfolio manager John Belton says the latest model releases from Meta and xAI confirm that Nvidia's competitive position in AI infrastructure remains strong, even as hyperscalers pour billions into developing their own custom chips. The analysis arrives at a moment when Wall Street is increasingly skeptical about the durability of Nvidia's revenue streams.
Companies like Meta, Alphabet, Amazon, and Microsoft - which collectively account for roughly half of Nvidia's business - are all investing in proprietary silicon designed to reduce their dependence on the chipmaker. Belton's read on the market suggests those efforts have not yet displaced Nvidia where it counts.
The frontier still runs on Nvidia
Belton pointed to Meta's Muse Spark 1.1 and xAI's Grok 4.5 as evidence. "Both models were trained on NVDA infrastructure, suggesting there is still a clear value proposition for using NVDA's stack," he wrote in a recent note. While in-house chip development continues to expand across the industry, the latest generation of frontier models shows Nvidia remains the platform of choice for the most demanding AI workloads.
Meta's spending plans extend the runway
Belton also flagged another signal supporting Nvidia's near-term outlook. Reports suggest Meta's AI infrastructure spending in 2027 could land well above current Wall Street estimates, indicating the hyperscaler capital expenditure cycle has further to run than many investors expect. That matters directly for revenue visibility, given that hyperscalers represent about 50% of Nvidia's business.
"The market has become concerned about the durability of those revenues," Belton wrote, noting that many of these companies are operating near break-even free cash flow. Meta's expanding ambitions provide greater near-term clarity into continued AI investment. For those tracking AI for Executives & Strategy, infrastructure spending patterns at this scale signal where competitive advantage is being built.
Fragmentation works in Nvidia's favor
Belton's most counterintuitive argument is that Nvidia does not need a single dominant AI winner. A fragmented market of competing AI labs is better for the chipmaker than a winner-take-all outcome. "Fragmentation in the LLM space is a good thing for NVDA," he wrote. If one company controlled the market for large language models, it could dictate terms to infrastructure suppliers. Instead, multiple well-funded competitors - Meta, xAI, Anthropic, OpenAI, and others - are racing to build better models, and each breakthrough drives additional demand for the hardware that makes training possible.
Why this matters for managers
The signal for decision-makers is straightforward: the infrastructure layer of AI remains concentrated even as the model layer fragments. Every major AI lab's progress depends on access to advanced computing resources, and that dependency is not shrinking. For managers evaluating technology strategy, assumptions about declining compute costs or the quick commoditization of AI infrastructure may be premature. Understanding these dynamics is central to AI for Management, where infrastructure dependencies directly shape build-versus-buy decisions and vendor negotiations.
Your membership also unlocks:
