AI Sector Shows Zero 'Dark GPUs' Amid $3 Trillion Infrastructure Boom, Signaling New Margin Realities

A recent discussion between David George, General Partner at a16z, and Gavin S. Baker, Managing Partner and CIO of Atreides Management, highlighted the profound economic shifts occurring within the artificial intelligence landscape. The conversation, held at a16z Runtime, delved into AI models, data centers, chips, and the critical implications of scaling laws and "The Bitter Lesson" for corporate margins. George shared on social media, "> Great to chat AI models, software, data centers, chips, bubbles, and everything AI with my good friend @GavinSBaker at @a16z Runtime. He is full of good takes, and here's his on what scaling laws and The Bitter Lesson mean for margins..."

Baker emphasized that the current AI investment boom fundamentally differs from past tech bubbles, such as the 2000 telecom crash. Unlike the "dark fiber" of that era, which saw 97% of laid cables unused, today's AI infrastructure shows "zero 'dark GPUs'," indicating active and full utilization of processing power. This intense demand is driving an estimated $3-4 trillion in planned data center expansion over the next five years, a scale surpassing the inflation-adjusted cost of the entire U.S. interstate highway system.

The core of Baker's analysis centers on how "scaling laws" and "The Bitter Lesson"—the principle that general methods leveraging increased computation and data outperform human-engineered solutions—are dictating AI company economics. These principles necessitate immense computational resources, leading to structurally lower gross margins for AI businesses compared to traditional software-as-a-service (SaaS) companies. Historically, SaaS firms enjoyed 80-90% gross margins, a figure AI companies are unlikely to match without significant changes in test-time compute requirements.

For investors and companies, this shift means that lower gross margins are becoming a "badge of honor," signaling genuine and deep engagement with AI technologies. Companies like Microsoft and Adobe successfully navigated margin compression during their cloud transitions, demonstrating that profitability can be sustained through increased revenue volume despite lower per-unit margins. This perspective encourages focusing on long-term positioning rather than dismissing AI opportunities based on immediate margin structures.

The discussion also touched upon the strategic "all-in" approach by tech giants like Google and Meta, viewing AI infrastructure investment as an existential necessity. Nvidia, a key player, is transforming from a semiconductor vendor to a comprehensive data center architecture company, integrating software, systems, and networking. This evolution, coupled with a move towards outcome-based pricing models for AI services, underscores the fundamental re-architecting of the technology and business landscape driven by AI.