Thursday, March 28, 2024

The Promise And Problem With NVIDIA’s AI Performance Benchmark Dominance

Datamation content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.

MLPerf is the definitive performance benchmark for Artificial Intelligence. The latest set of competitive performance benchmarks were released this week, and NVIDIA so dominated the benchmarks they effectively stand alone in this market right now for core functions like image classification, object detection, and machine translation. Others either didn’t have a technology ready to submit to a category, or their performance was so low to be arguably meaningless. 

This lead provides NVIDIA with both a substantial competitive advantage and a significant problem. This result is because it is hard to have a race when only one team is competitive and, in the early phases of a market (and we are still clearly in the early phases) you need a lot of viable players and, at least in the segments measured by MLPerf there is only one, NVIDIA. 

Let’s talk about the promise and the problems with being so far out front.

NVIDIA’s AI Promise

By being so dominant in this space, it places NVIDIA’s core technology at the top of any shortlist for those partnering in this area. It helps explain why Mercedes, a famous company for not accepting technology from others for critical strategic programs, partnered with NVIDIA. If they wanted a viable self-driving platform backed by deep simulation, NVIDIA was the only short term path to achieving that goal. 

As a result, expect NVIDIA to gain even more partners, to dominate more industries, and to remain the definitive leader of the segment.  Tactically they are unbeatable.  But this kind of overwhelming leadership can lead to problems.

There are three unique problems that NVIDIA will now have to deal with, given their extreme leadership.  They are false confidence, regulatory oversight, and higher than average intellectual property theft.  Let’s take each in turn.

False Confidence

While most every major AI tech vendor (and several universities) is participating in MLPerf, it is unusual to submit results when you are well off the mark.  Those not submitting benchmark results could be far better off than these latest charts indicate, and they may be far more competitive than they appear as a result.  

NVIDIA, believing they are unbeatable, could falsely feel their lead is unbeatable and slow down their impressively fast development effort allowing the other firm(s) to pass them and then be unable to keep up. While there is no evidence I can see that NVIDIA is slowing down, the risk remains very real. A closer ranked competitor generally will help drive the leader to ever-higher performance, something that NVIDIA will have to do itself until such a competitor emerges. 

Regulatory Oversight

When you are effectively the only vendor capable of performing at an adequate level, you own the market.  That could have anti-trust implications because of the resulting market power. We just saw Apple, Amazon, Facebook, and Google begin what will likely be a dance concluding with one or more of them being broken up (odds favor Google and Amazon being broken up, Facebook being sanctioned, and Apple being left to the States which have begun their actions). 

Tighter competition generally shields the leading company from this kind of government response. Right now, NVIDIA doesn’t appear on any anti-trust action I’m aware of, but NVIDIA will need to be very careful they don’t abuse their market power to avoid being in that hot seat.  NVIDIA has a decent reputation with its partners, and if they maintain that, they should be safe here, but it remains a risk. 

Intellectual Property Theft

When you are as far out ahead as NVIDIA does competitors, and particularly other countries will look to mine you for information illegally. This means everything from placing people in the company to get that information, appearing as customers rather than IP thieves to gain access to it that way. 

This level of leadership in a technology area seen as a critical part of the next industrial revolution will be an extremely attractive lure that many may not be able to resist.  NVIDIA needs to be able to share its technology with customers (it has sizeable open-source component) so that those customers can implement it in their products but, in so doing, will increase the likelihood of IP theft, meaning the firm will need to be extraordinarily vigilant.   

NVIDIA Once Again Demonstrates They Are The Company To Beat, But….

When it comes to core AI technology, NVIDIA remains the company to beat and by an impressively significant margin. However, this is based on public information, and few are sharing where they are. This can lead to several problems for NVIDIA ranging from false confidence to intellectual property theft that NVIDIA should be able to deal with if they don’t take their eyes off the ball, and NVIDIA’s CEO is known for having eyes in the back of his head. 

The biggest potential problem, though, is the abuse of their market power because that is something that tends to catch every company that becomes dominant up eventually and it is the most difficult to mitigate because it means putting in place oversight which managers universally find annoying. 

Right now, NVIDIA stands alone, that will change, and whether NVIDIA remains out front or not, as this market matures, a lot will have to do with how the firm anticipates the related unique threats, including typical executive behavior. 

Subscribe to Data Insider

Learn the latest news and best practices about data science, big data analytics, artificial intelligence, data security, and more.

Similar articles

Get the Free Newsletter!

Subscribe to Data Insider for top news, trends & analysis

Latest Articles