Nvidia (NASDAQ: NVDA) has long been the dominant force in artificial intelligence chips. Its best-in-class GPUs (graphics processing units) provide a combination of flexibility and performance that’s unmatched by the offerings of any of its rivals. Perhaps its biggest advantage, though, is the CUDA programming interface that translates code into instructions for its processors — and only its processors.
For years, developers have built frameworks that rely on CUDA and Nvidia’s architecture, which has essentially locked many data centers into using Nvidia’s chips.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a “Double Down” signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same “Total Conviction” signal is flashing for a company 1/100th the size of Nvidia. Continue »
But Alphabet (NASDAQ: GOOG) (NASDAQ: GOOGL), Meta Platforms (NASDAQ: META), and Oracle (NYSE: ORCL) could disrupt that massive competitive advantage with several new developments. If investors aren’t paying attention, they could overlook a couple of key details.
Can Alphabet take on Nvidia?
One of the most common development frameworks for AI code today is PyTorch, originally developed by Meta Platforms. Meta uses PyTorch for its deep learning and large language model development, and the open-source framework is standard across the industry. Importantly, PyTorch is optimized for Nvidia’s CUDA with direct integration with Nvidia’s libraries and hardware features. That has strengthened the lock-in for Nvidia’s chips.
Developers who want to run their software on a different chip need to port their code to a different hardware interface, which typically requires extensive changes. But that changed recently. Alphabet announced TorchTPU last month, an open-source tech stack that lets developers run PyTorch workloads on its TPUs (Tensor Processing Units) with minimal code changes. Amazon, likewise, announced its own backend that supports PyTorch code on its Trainium chips.
Meta has worked with both to ensure it can use competing AI accelerator chips and avoid being compelled to stick with Nvidia’s.
“At Meta, advancing the frontier of AI requires immense compute scale and deep hardware diversity. We cannot be constrained by a single hardware architecture,” said Jana van Greunen, Meta’s director of PyTorch engineering, in Alphabet’s press release.
Indeed, Meta faces the same challenges as everyone else developing frontier AI models: Because demand for Nvidia GPUs exceeds supply, access to them is limited and expensive. These recent software-layer developments could ease that challenge and eventually result in Nvidia losing a significant amount of its now-dominant market share in the data center chip segment.
It’s important to note that just because PyTorch can easily run on Alphabet’s TPUs and Amazon’s Trainium chips now, that does not mean it runs as well on them as it does on Nvidia’s GPUs. As mentioned, PyTorch is still optimized for CUDA and Nvidia’s hardware. That said, the recent developments open the door for further optimization and ease of portability, and the competing chips could become much more performant in the future.
Alphabet is bringing its chips to more developers
Developers interested in using Alphabet’s TPUs for AI training or inference previously had to work directly with Alphabet’s Google Cloud Platform for access to those chips. That’s starting to change, though. Earlier this year, Alphabet started selling TPU systems to select third parties, and it’s ramping sales up quickly. Still, access to TPU systems remains relatively tight.
But Alphabet announced a deal with Oracle at the end of last month that will bring its TPUs to Oracle Cloud Infrastructure starting in 2027. The move gives developers that already work with Oracle the opportunity to deploy some workloads on TPUs, which could offer better performance, pricing, or both relative to Nvidia’s GPUs. What’s more, it could provide developers locked into Oracle because they host databases with the cloud provider an opportunity to use TPUs.
The move by Oracle to spend some of its capital on Alphabet’s chips instead of Nvidia’s or other GPU makers is notable. It reflects the potential for cloud providers to generate strong returns on invested capital from competing AI accelerators, which will further weaken Nvidia’s position. With the growing utility of custom chips from Alphabet and Amazon, among others, thanks to PyTorch developments, the position is further weakened.
Nvidia’s stranglehold on data center spending is eroding, and these recent developments suggest its competitive moat might not be as wide as some investors previously thought. With competing chips offering lower costs and growing capabilities, it could prove challenging for Nvidia to continue growing at the rapid pace investors expect. That said, its stock valuation has reflected that potential risk for some time: Shares today trade at about 22 times forward earnings expectations. However, even at that price, there are other AI stocks I’d rather own, including Alphabet, which has made tremendous progress with its own chips.
Should you buy stock in Nvidia right now?
Before you buy stock in Nvidia, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Nvidia wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004… if you invested $1,000 at the time of our recommendation, you’d have $341,966!* Or when Nvidia made this list on April 15, 2005… if you invested $1,000 at the time of our recommendation, you’d have $1,146,819!*
Now, it’s worth noting Stock Advisor’s total average return is 852% — a market-crushing outperformance compared to 203% for the S&P 500. Don’t miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
https://finance.yahoo.com/technology/ai/articles/alphabet-meta-oracle-teaming-disrupt-101500132.html?shem=aimgspe,i3ppal,
