The AI Revolution's $600 Billion Question: Are We Overinvesting in GPUs?

The artificial intelligence (AI) landscape is rapidly evolving, with advancements happening at an unprecedented pace. But beneath the surface of groundbreaking innovations and soaring valuations, a crucial question looms: Are we building an AI infrastructure that outstrips the current demand and revenue generated by AI applications?

This article delves into the financial intricacies of the AI boom, exploring the potential gap between infrastructure investment and actual revenue growth in the AI ecosystem. We will discover the key challenges and opportunities that lie ahead.

Decoding the $600 Billion Hole

In September 2023, concerns were raised about a potential "$200 billion question" in the AI space. This figure represented the revenue needed to justify the massive investments in AI infrastructure, particularly Graphics Processing Units (GPUs). Fast forward to today, and that question has ballooned into a staggering $600 billion question.

To understand this figure, consider the following calculation:

1. Nvidia's Run-Rate Revenue: Take Nvidia's projected revenue.
2. Total Data Center Costs: Multiply it by 2 to account for the total cost of AI data centers (GPUs typically make up half the cost).
3. Gross Margin: Multiply again by 2 to factor in a 50% gross margin for the end-users of GPU computing, such as startups and businesses using cloud services like Azure, AWS, or GCP.

Key Changes Since 2023

Several factors have contributed to this widening gap:

• Subsiding Supply Shortage: The GPU shortage that plagued late 2023 has largely dissipated.
• Growing GPU Stockpiles: Major cloud providers like Microsoft are heavily investing in GPUs, leading to significant stockpiles. Microsoft alone accounted for roughly 22% of Nvidia's Q4 revenue. However, stockpiling can lead to decreased demand once those stockpiles are large enough
• OpenAI's Dominance: While many AI startups are emerging, OpenAI, with its ChatGPT, still captures the lion's share of AI revenue. In fact, OpenAI's revenue is now $3.4B.

The B100 Factor: Fueling Further Demand

Nvidia's announcement of the B100 chip promises 2.5x better performance. This is expected to further amplify the demand for NVDA chips, potentially creating another supply shortage as companies race to acquire the latest technology.

Comparing AI Infrastructure to Railroads

One argument suggests that GPU CapEx is similar to building railroads. Eventually the 'trains' will come, bringing with them new economic growth. While this analogy holds some truth, there are critical differences:

• Lack of Pricing Power: Railroads possess inherent monopolistic pricing power due to the limited availability of tracks between locations. In contrast, GPU computing faces increasing commoditization, with numerous new entrants flooding the market with dedicated AI clouds.
• Investment Incineration: Speculative technology waves often lead to significant capital incineration.
• Depreciation: Semiconductor technology is constantly improving, leading to rapid depreciation of older chips as Nvidia and others introduce newer, more powerful GPUs.

These factors suggest that overestimating the long-term value of current-generation GPUs is a real risk.

Winners and Losers in the AI Gold Rush

Despite the potential for overinvestment, the AI revolution will undoubtedly create winners. Declining GPU computing prices can be beneficial for long-term innovation and startups. Even in the case of overinvestment, founders, and company builders can thrive by leveraging lower costs and learnings.

Avoid the Delusion and Build for the Long Term

It's crucial to avoid the delusion of quick riches driven by the anticipation of imminent Artificial General Intelligence (AGI). Stockpiling GPUs in the hopes of striking gold overnight is a risky and unsustainable approach.

The AI revolution is a marathon, not a sprint. Success will come to those who remain level-headed, focus on delivering value to end-users, and build sustainable businesses for the long term.

Key Takeaways:

• The AI infrastructure build-out may be outpacing current revenue generation, creating a significant financial gap.
• GPU stockpiling and the lack of pricing power in GPU computing pose challenges to long-term profitability.
• Focusing on delivering real value to end-users and building sustainable businesses is crucial for success in the AI landscape.

If you're building in the AI space, remember that the road ahead is long and filled with challenges, but the potential rewards are immense.