Business

Booming AI Demand Outpaces Fears of Frontier Model Slowdowns

Despite recent industry-wide hand-wringing over the pace of frontier artificial intelligence development, major chip manufacturers continue to report record-breaking financial growth, suggesting that any temporary slowdown in model training has done little to curb demand for computing hardware.

Throughout the summer and early fall of 2026, concerns emerged that the most powerful AI models might be hitting a plateau in development. In August, OpenAI paused its primary frontier training run after its AI agents inadvertently hacked Hugging Face, with early indications suggesting the next generation of models could independently identify and exploit security vulnerabilities. Shortly thereafter in September, Anthropic CEO Dario Amodei publicly advocated for slowing the development of advanced models, citing risks related to potential cyberattacks and difficulties in maintaining control over such systems.

image

While these concerns triggered a short-term sell-off in U.S. chip stocks, the anticipated cooling of the market has not materialized. OpenAI restarted its training run after a brief two-week pause, and by November, Anthropic had released upgraded versions of its Fable and Mythos models.

Financial results from leading semiconductor companies underscore the continued surge in demand. Nvidia announced a blockbuster quarter on Tuesday, with data center revenue climbing 110 percent year-over-year to $107.6 billion for the period ending October 25. The company noted that its new Vera Rubin NVL72 computing system represents the fastest product ramp in its history, generating nearly $20 billion in revenue during its initial quarter of production. Interim CEO Colette Kress stated that Nvidia expects major frontier laboratories to represent approximately 15 percent of its data center revenue this fiscal year and subsequently raised the company’s growth forecast for the next fiscal year from 70 percent to 75 percent.

image

Similarly, AMD reported its strongest quarter to date earlier this month, with total revenue rising 45 percent to $13.4 billion, largely driven by data center sales which nearly doubled to $8.5 billion. OpenAI and Anthropic are currently among the primary users of AMD’s high-end Helios systems.

Industry analysts suggest that even if labs were to significantly scale back training, the financial impact would likely be delayed by existing procurement agreements. Neil Shah, vice president of research at Counterpoint Research, noted that while a shift toward inference—the process of running pre-existing models—might appear to reduce the immediate need for top-end training GPUs, it creates a net increase in demand for total compute infrastructure. According to data from the corporate spending platform Ramp, frontier model usage among businesses reached a record high in early November, accounting for approximately 60 percent of token usage.

image

However, experts observe that the market may be evolving toward more efficient hardware. As AI companies focus on the economics of serving models to large user bases, they are increasingly exploring custom chips. Counterpoint Research projects that by 2028, shipments of custom AI server processors may surpass those of general-purpose data center GPUs. While UBS analyst Timothy Arcuri questioned whether Nvidia would eventually sacrifice profit margins to compete with these custom solutions, the company maintains that its systems justify their current price points.

Charlotte Hayes

Charlotte Hayes writes about economic policy, inflation, employment, trade, and financial trends. Her coverage connects economic data with everyday consequences, giving readers clear explanations of the forces affecting households, businesses, and national economies.
Back to top button