In the face of Google's challenge, who is more vulnerable, NVIDIA or OpenAI?

Author: Ye Huiwen

Source: Hard AI

On Monday, December 1st, senior technology analyst Ben Thompson published an article on his widely followed blog Stratechery, delving into the current competitive landscape in the AI field, likening the development of the entire AI industry to a classic "hero's journey" epic.

Thompson, known for his profound business strategy analysis, believes that in the narrative of AI over the past few years, OpenAI and NVIDIA are undoubtedly the two main protagonists.

One has jumped from a startup to a phenomenal consumer technology company, while the other has transformed from a gaming chip manufacturer to the cornerstone of the AI revolution's "arsenal." However, just as the hero Luke in the movie "Star Wars" must enter the "Death Star" to face his trials, these two companies are now facing their "Empire Strikes Back" — Google is launching a thunderous counterattack.

In the face of Google's strong counterattack as a former dominant player, which of the two main protagonists in the AI field over the past few years — hardware giant NVIDIA and model leader OpenAI — is in a more dangerous position?

Thompson's analysis points directly to the foundations of the business models of both companies. He believes that while OpenAI continues to "burn money," and NVIDIA is "printing money" like crazy, from the fundamental attributes of their moats, OpenAI's advantages may be more solid than NVIDIA's.

However, he also sharply points out that OpenAI itself seems to be unaware of this, even actively undermining its greatest advantage, which poses a significant hidden danger for its future.

Google's Counterattack: A Two-Front Battle

First, let's take a look at how Google is "counterattacking." Thompson notes that Google's first move was to release its powerful Gemini 3 model. This model has surpassed OpenAI's most advanced models in several benchmark tests, proving that Google still possesses unfathomable strength in technological research and development. This directly undermines OpenAI's foundation as the "best model provider."

More critically, Google's second move targets NVIDIA's lifeblood.

In the past, it was generally believed that even if Google's self-developed TPU chips had superior performance, they were only for internal use. But now the situation has changed; Google has begun to sell TPUs as alternatives to NVIDIA's GPUs in the market and has already reached cooperation intentions with giants like Anthropic and Meta. This is akin to a powerful competitor suddenly appearing in NVIDIA's profit-rich backyard.

As Thompson puts it, this has led "NVIDIA to suddenly realize that it is also in the eye of the storm, with external parties questioning the sustainability of its long-term growth, especially its exorbitant profit margins."

NVIDIA's Moat: Seemingly Solid, Yet Concealing Cracks

So, is NVIDIA's moat deep enough? Thompson analyzes that NVIDIA's advantages mainly consist of three points: superior performance, greater versatility (GPUs are more flexible than TPUs), and a strong developer ecosystem built on the CUDA software platform. However, as Google's TPU performance catches up or even surpasses, the first advantage is weakened The deeper crisis lies in the CUDA ecosystem. Thompson used a very clever analogy: how AMD made a comeback in the data center market against Intel. He pointed out that it was the large-scale cloud service providers like Google and Microsoft that found it "worth it" to invest resources in rewriting the underlying software to be compatible with both AMD and Intel chips, thus breaking Intel's monopoly.

Today, NVIDIA faces the same problem, its customers are highly concentrated among a few giants, and these giants have every motivation and resource to "tear down the wall of CUDA." Thompson quoted his past article saying, "The pressure and possibility of escaping CUDA are higher than ever." Although NVIDIA's position is difficult to shake in the short term, the long-term risk of eroded profit margins has already emerged.

OpenAI's Ace: The Choice of 800 Million Users

In contrast, what about OpenAI's situation? Although it appears financially weaker, Thompson believes its moat is fundamentally different. NVIDIA's customers are a few large companies, while ChatGPT's territory is a vast consumer market with over 800 million weekly active users.

Here, Thompson made a simple yet profound point: “ The solidity of a moat is proportional to the number of independent users.”

Why is that? He explained that getting a CEO of a large company to order a change in the tech stack might only require a few meetings and a nice PowerPoint presentation. But changing the daily habits of 800 million users is a "street battle" that needs to be fought one by one.

This network effect, arising from spontaneous consumer choice, is the hardest barrier to breach. Thompson sharply pointed out: “Changing the habits of over 800 million people who use ChatGPT weekly is a battle that can only be fought on an individual basis. This is the real difference between ChatGPT's battle against Google and NVIDIA's situation.”

The Biggest Concern: Holding an Ace but Not Knowing How to Play It

Since it has such a strong user base, why is OpenAI still considered at risk? This leads to Thompson's most critical criticism in the article. He believes OpenAI has made a huge business mistake: it has yet to launch an advertising model.

He pointed out sharply that for an aggregator platform with a massive user base, advertising is not only the best monetization model but also a catalyst that can make the product better Advertising can attract more free users, bringing in more usage data and feedback to optimize the model; at the same time, by capturing users' purchasing intentions, it can lead to a deeper understanding of users and provide more personalized services.

As another expert, Eric Seufert, pointed out, Google began monetizing ads less than two years after launching its search engine, and it was the advertising revenue that supported all its subsequent innovations. Thompson bluntly stated that OpenAI's adherence to a subscription model three years later is a form of "commercial negligence," especially given its commitment to purchasing trillions of dollars in computing power. This approach is tantamount to handing over the vast free user market to Google, which excels in this area.

The Ultimate Test of Business Models

In the end, Thompson elevated this competition to the ultimate test of his own "Aggregation Theory." He has always believed that in the internet world, whoever can control user demand holds the ultimate power. The rise of ChatGPT perfectly confirms this point.

However, the current question is whether an already successful aggregator (Google), with its overwhelming resource advantages, can defeat an emerging challenger (OpenAI) that has not fully leveraged its aggregation advantages (i.e., the advertising model). Thompson admitted that he feels both nervous and excited about this.

In summary, Thompson's core argument is: NVIDIA's vulnerability lies in its business model, as its high profits depend on a few major clients capable of "betraying" it; while OpenAI's advantage lies in its large user base, but this advantage is being undermined by its shortsighted business model. The ultimate outcome of this battle among giants will not only determine the fate of these companies but may also redefine the fundamental rules of competition among future technology platforms: is vast resources more important, or is extreme control over user demand more advantageous? This is undoubtedly one of the most noteworthy aspects of the technology industry in the coming years