Boss Digital


Opinion: Power Politics and GPUs

Editor’s take: As much as it would make sense for Nvidia to focus solely on being the leading AI silicon vendor, their rise to power has left them with little choice but to continue pushing forward in areas that make some of their big customers uncomfortable.

The big news in the world of tech last week came from the Computex trade show in Taiwan. The event came with the usual slew of press releases, keynotes, and of course, the continued rise of Nvidia CEO Jensen Huang’s rockstar status, which we are now apparently calling Jensanity. (Someone should study the timespan between a CEO achieving celebrity status and the surrounding bubble bursting. It cannot be a good sign).

There has been a lot of good coverage of the show, but Ben Thompson’s piece stood out to us. Thompson argues that Huang’s Computex keynote expresses the optimal strategy for the company: focusing on designing the best chips and supplying the infrastructure for the AI boom revolution momentum. This contrasts with past keynotes, where Huang seemed to emphasize the company’s software and cloud ambitions.

Editor’s Note:
Guest author Jonathan Goldberg is the founder of D2D Advisory, a multi-functional consulting firm. Jonathan has developed growth strategies and alliances for companies in the mobile, networking, gaming, and software industries.

This strikes us as a highly sensible assessment. The company has an obvious core competency in designing these chips and a massive head start on one of the biggest shifts in compute spending in memory. Focusing on that is almost certainly the smart move at this point. Going too far down the software path risks distractions, lost focus, and the threat of ending up in competition with their largest customers. Focusing on chips is the best strategy. Totally makes sense.

That being said, we have to wonder if it is already too late for the company to hew to this path. We are over four years into Nvidia’s rise to prominence in data center silicon and well over a year into the post-ChatGPT surge in Nvidia’s revenue.

They have already made it clear they have a lot of software in the works, and that software, in the form of CUDA, was a big part of their competitive advantage. Those big customers, the hyperscalers, are all keenly aware of their dependence on Nvidia and the fact that they cannot get nearly as many Nvidia GPUs as they want. Put simply, the power dynamic has already shifted.

I’ve been listening to a lot of history podcasts lately, and one clear lesson from those is that good intentions do not matter when there is power at stake. Ludwig of Bavaria and Henry of Austria may have been raised together, best friends all their lives, but at some point, their two kingdoms went to war, and they had to try to kill each other.

The war for AI has already started, and even if Nvidia “just” wanted to be the leading AI semiconductor vendor, their range of options is now limited by the state of the market and the perception of them held by the other combatants.

There are signs of this already playing out. The hyperscalers are all designing their own AI accelerators, and they all say (implicitly, but often explicitly) that part of their motivation for doing this is to reduce their reliance on Nvidia. The company may now try to tell everyone they are good faith silicon vendors working crazy hours to make sure everyone gets the chips they want, and that their software is non-competitive with the hyperscalers, just a nice added feature of their hardware. They may try this, but it is unlikely the hyperscalers will believe them; paranoia is a watchword in the Valley.

And even if the company were sincere in this message, they cannot actually deliver what the big customers want. Nvidia does not have enough chips to meet demand; everyone is on allocation. That means Nvidia holds the power in all negotiations, whether they want it or not.

And of course, they do want that power. They would rather sell complete systems with complete boards, fully-marked-up memory, and pricey networking. How are they actually deciding who gets allocated how many chips? They would have to exert superhuman amounts of self-discipline to not favor the customers who pay Nvidia more, and that level of discipline is probably in conflict with their duty to shareholders.

The hyperscalers are also pushing very hard to dilute Nvidia’s software barriers to entry. It is not clear they can do this, but they likely see it as critical that they push as hard as they can. And how will Nvidia respond to this? They could just announce they are exiting software, but that would be nearly suicidal and would not really accomplish much. Instead, they are going to have to move in the opposite direction and double down on their investment in software to shore up their differentiation. Which, of course, just reinforces the cycle.

Throughout this, good intentions are at best meaningless. The power dynamics at play leave the company with a limited set of options for moving forward. To be clear, they have a lot of great options, their position is immensely strong, but we do not think they can move back to being “just” a semiconductor supplier. And it is not clear to us that they even want to do that.

Source link

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top