Ben Thompson, founder of Stratechery and a renowned technology analyst, recently participated in an in-depth discussion. Drawing on two decades of experience observing the tech industry, Thompson offered sharp commentary during the 90-minute interview on computational bottlenecks in the AI era, the evolution of business models, and the moats of tech giants.
A core warning from Thompson centers on the conservative approach of Taiwan Semiconductor Manufacturing and a potential widespread chip shortage by 2029. As the proponent of Aggregation Theory, he expressed significant concern about the current pace of AI infrastructure development. His central argument is that the limiting factor for global AI expansion is actually the speed of TSMC's capacity expansion.
Thompson pointed out that despite enormous market demand, TSMC, as a near-monopolist, has been quite conservative in expanding production. This is because building fabrication plants carries extremely high risk; if overcapacity occurs, massive depreciation costs could decimate profit margins.
"99.9% of a fab's cost is depreciation... TSMC is actually behaving rationally. They would rather forgo potential long-term revenue than shoulder the downside risk of overcapacity," Thompson explained.
This conservative strategy creates a risk mismatch. TSMC transfers the risk of insufficient capacity to companies like Nvidia, Apple, and major cloud providers. These tech giants, in turn, face the risk of "losing future revenue due to a lack of computing power."
Thompson made a striking prediction: "I believe we will face a massive chip shortage around 2029."
He emphasized that the current increases in capital expenditure are still insufficient to meet the exponential growth in computational density expected from future AI agents. He urged hyperscalers to support competitors like Intel or Samsung—driven by pure economic motives rather than just geopolitical considerations—or to share the risk of building new fabs through pre-payments. Failure to do so, he warned, could leave them trapped by a production bottleneck.
On the topic of AI business models, Thompson challenged the Silicon Valley consensus. While the tech community often views advertising with skepticism, he firmly believes it is the most effective monetization form for AI, especially for companies like OpenAI that have massive user traffic but lack a clear commercial loop.
He refuted the idea that "advertising degrades the quality of AI answers," citing a key insight from Mark Zuckerberg: "Zuck once said that the largest, most successful Agent in the world today is actually the Facebook ad system."
Thompson elaborated that Facebook's ad system is essentially an automated agent: a business inputs a budget and target, and the system automatically finds customers and delivers results. This, he argues, is the ultimate form of an agent. Regarding OpenAI's current attempts to place ads based on conversation context, he criticized the approach as likely to irritate users.
"The best advertising model isn't based on what you're currently chatting about, but on a comprehensive understanding of you as a person. Google has a huge advantage here; they don't need to stuff ads into the Gemini dialog box. They can use Gemini's data to monetize more accurately on YouTube or Search."
Thompson also delivered pointed assessments of the major US tech giants, the "Big Five":
Meta was praised for having the strongest execution. Despite market concerns about its substantial capital expenditures, Thompson believes Meta's advertising model is undervalued. He noted that Meta not only has strong cash flow but is also building infrastructure through its open-source models. "The threat from OpenAI might be greater to Facebook than to Google, but Facebook is clearly spending money to meet this challenge," he added.
Google was described as chaotic yet resilient. Thompson compared Google to a slime mold: "It seems messy, suboptimal, with lots of redundant actions, but this lack of optimization actually gives it tremendous adaptability and resilience. When it finally moves toward you, you're finished."
Amazon's strategy was flagged as risky. Thompson expressed concern about Amazon's chip strategy in the AI era. While Amazon is accustomed to winning in commoditized markets through low costs, this approach might fail in AI, where each new generation offers a massive performance leap. Relying on its own Trainium chips instead of the most powerful Nvidia chips could lead to a competitive disadvantage.
Apple was critiqued for its platform management. While its hardware is considered unbeatable, its software and services platforms are poorly managed. "Apple makes great products but are terrible stewards of platforms," Thompson stated.
Looking at the broader software industry, Thompson suggested that the SaaS business model, which often charges "per seat," could face a growth ceiling if AI leads to reductions in company headcount.
Finally, regarding a future flooded with AI-generated content, Thompson argued that scarcity will redefine value. "In a world of infinite content, anything 'live' becomes more valuable. Shared experiences, in-person classrooms, sporting events—these communal experiences that cannot be personalized by AI will be where the premium lies in the future."
Comments