Is Nvidia the Intel of the 2000s? | Sept ‘25

Thoughts are welcome via emails to zhenghan236@gmail.com, and will be featured in the next Tab

---

As someone who shares both a surname and house address as CEO Jensen Huang, I would expectedly be a fervent supporter of his star-studded firm. Yet my experience with Nvidia products has been surprisingly limited. The reasons are simple: I am a Mac user (the Apple-Nvidia feud is a fascinating saga worth reading about), I don’t really play video games, and I never had the patience to dig up internet tokens with a virtual shovel — however valuable they turned out to be.

While I’ve remained at arm’s length from Nvidia GPUs, the rest of the finance world seems to think otherwise about its maker. Case in point: since ChatGPT’s public debut in fall 2022, galloping sales of Nvidia’s AI processors have propelled its stock price by more than 1100%. This summer, it even unseated Microsoft to become the world’s most valuable company. For a stock trading at 48 times earnings, its financial footing appears to rest on remarkably solid bedrock. But why is that? After all, Nvidia produces intermediates — chips; no one is going to care if their ChatGPT response was processed on an Nvidia chip.

What data centres do care about, however, is performance, and that’s where Nvidia has built its moat. At the heart of this advantage is CUDA (Compute Unified Device Architecture), a proprietary low-level platform that lets developers tap into the parallel computing power of Nvidia GPUs by treating them like super-parallel CPUs. Nvidia does not just sell chips, it sells the whole stack of hardware, CUDA drivers, libraries, and developer tools. It was early to recognise that high-performance computing (which includes AI) required parallel compute, and by the time AMD and Intel tried to catch up, CUDA had already become the default language. Today, CUDA supports over 80% of GPU-accelerated AI workloads and a developer base of over 4 million worldwide. In a sense, that has enabled Nvidia to position itself more like an “End Firm,” with avenues to differentiate and lock in customers.

Sixty percent operating margins thus seem realistic for Nvidia. More impressive still is its astounding growth: revenues are up 114% y/y while cash holdings have swelled to over $60B — an uptick of 125% over the same timeframe. Its working capital efficiency is also superior to its peers Intel and AMD, with Days of Inventory Outstanding of 86 days compared to 119 and 152 respectively. Days Payable Outstanding for Nvidia is also, well, outstanding, at 50 compared to 60 and 107. This alludes to Nvidia’s superior operational efficiency and significant leverage over suppliers.

Yet fundamentally, Nvidia faces some long-term headwinds. Like Intel’s loss of dominance over x86, Nvidia’s CUDA monopoly is not bulletproof — especially if the industry shifts away from that architecture (Broadcom’s ASIC based approach comes to mind). Complacency led Intel to miss the ARM revolution in mobile computing, while Apple circumvented the x86 lock-in entirely with its Rosetta translation tool. In fact, the biggest buyers of Nvidia GPUs — Microsoft, Google, and AWS — are already designing their custom chips to reduce dependence and optimise performance. Apple, meanwhile, runs Private Cloud Compute on Apple Silicon. What guarantees does Nvidia have that it can withstand similar shifts?

It also faces the risk of demand cyclicality. Nvidia has historically maintained its edge by riding successive computing waves. It did so with GeForce and gaming in 1999, crypto mining in the 2010s, and most recently, AI. Unsurprisingly, the dot-com bust in 1999 caused Nvidia GPU demand to crash. The 2018 crypto implosion led to an inventory glut. Given its immense reliance on AI, should the bubble burst, Nvidia may find itself the largest casualty of the debacle without a contingency plan. In a sector of fast obsolescence like technology, Nvidia’s technology-based moat, like Intel’s, is wide but dynamic and shallow.

As I finish this piece on my too-hot-to-handle Intel MacBook Pro, fully conscious that Apple Silicon exists, I can’t help but wonder if today’s AI giants may one day be left with a sour taste — watching their shiny new in-house processors match Nvidia’s flagships at half the cost. Just as Intel’s x86 empire was toppled by new architectures and customer-driven revolts, Nvidia’s CUDA dominance could face the same reckoning. Whether it becomes the Intel of the 2000s — a fallen giant — or rewrites the script depends on how long AI remains its golden wave.