Nvidia reports results: are fears of a bubble really justified?
20 NOV, 2025
By Joanna Piwko from RankiaPro Europe
With recent market angst around circular financing and an AI bubble it was an opportune moment for Nvidia to set the record straight on what they were seeing from their vantage point as the AI leader. Jensen Huang gave an impassioned defence of why 3 vectors were driving a huge investment in new compute infrastructure that still in his view had many years of growth to run:
- The End of Moore’s Law: existing compute infrastructure that underpins all of our data and software no longer scales given traditional Moore’s Law has slowed driving the need to transition to accelerated compute to drive down cost.
- Generative AI: the internet services we interact with every day are built on search rankings, recommender engines and ad targeting. Large language models are much more effective in driving these engines, which in turns creates more revenue via more engagement and traffic and better identifying commercial intent leading to higher click through rates.
- Agentic AI/Physical AI: these new revolutionary technologies are creating brand new, very large addressable markets. This is driving some of fastest revenue ramps we have ever seen with Jensen referencing Anthropic going from $1bn in annualised revenue earlier this year to $7bn annualised last month.
Other recent concerns have been Michael Burry’s short thesis around hyperscalers overstating earnings via understating depreciation, circular financing and recent customer deals with competitors.
- Useful Life of A GPU: the CFO made the point that 6 year old Ampere GPUs are still available in the cloud and fully utilised today in contrast to Michael Burry’s assertion that the useful life is really only 2-3yrs. While you would not use a 6 year old GPU to train a frontier model, you can use it for less intensive workloads while they also made the point that the performance of an old GPU does not sit still as software upgrades can drive meaningfully improved performance over time (4-5x for Hopper over its lifetime to date).
- Circular Financing: recent announced investments by Nvidia in OpenAI and Anthropic need to be put in context of a company that generated $22bn in free cashflow last quarter, has over $50bn of net cash on the balance sheet and is forecast to generate an additional $300bn+ in free cashflow in the next 2 years alone. Nvidia defended these deals as a way to expand their ecosystem with Anthropic committing this week to use Nvidia GPUs for the first time, accelerating the growth plans of these new customers as well as the potential future financial returns of taking stakes in companies they view as being generational in their capabilities.
- Competition: Nvidia made the point that their proprietary programming language CUDA as well as the versatile architecture allow their GPUs to have a much longer life versus the competition that is limited to a few years as model technologies evolve. It is worth noting that while AMD at their recent analyst day promised tends of billions of dollars of AI compute revenues in 2027, Nvidia just reported that in one quarter. While others are promising jam tomorrow, Nvidia is jam today.
Back to the results themselves, Nvidia delivered more than the usual quantum of upside versus expectations, re-accelerating growth and having already laid out at their recent GTC Washington event a pathway to over $300bn in datacentre sales next year they said this week’s HUMAIN Saudi Arabia deal as well as Anthropic were incremental to that number. That continues to include de minimis China sales given the ongoing impasse but Nvidia has not given up hope, emphasising the need to have a competitive product to sell in China and the importance of America being able to compete globally. Buyside forecasts continue to be well ahead of formal sell-side consensus so the upgrades overnight are more of a catch up but the stock continues to trade at a far from egregious valuation, which remains a key retort to recent AI bubble concerns and comparisons with 2000.
