You’re Getting AI Wrong | GMO’s Tom Hancock on Finding Conviction Amid the Hype

Applications layer remains uncertain

While code generation and chatbots are proven monetizable applications, most future killer apps remain unidentified, making it premature to pick winners among startups versus legacy players with existing data and customer lock-in.

Hyperscalers provide essential compute

Microsoft Azure, Google Cloud, and AWS host current AI workloads and will likely power future applications, positioning them as the crucial middle layer regardless of which specific applications ultimately dominate.

LLMs occupy the innovation center

Large Language Models represent the core technical breakthrough but are concentrated in private companies or big tech subsidiaries with unclear long-term differentiation or economic moats.

Infrastructure suppliers lack demand visibility

Nvidia and semiconductor suppliers sit at the bottom of the stack with high current revenues but the least visibility into ultimate end-demand, making their sales entirely dependent on capex decisions from layers above.

Revenue flows downward through the stack

Nvidia's revenue equals OpenAI's capex, which is funded by Microsoft and outside investors, meaning supplier revenues are only as durable as the spending appetite of the application and hyperscaler layers above them.

Strategic cash reduces bubble risk

Unlike the 1999-2000 tech bubble funded by debt and equity issuance, current AI investment is largely funded by Big Tech's internal cash flows, providing stability against interest rate shocks or capital market closures.

Finite reserves signal future constraints

While Microsoft and Alphabet maintain deep pockets, some players are approaching break-even cash flow levels, suggesting future investment may require debt issuance or slowed growth rates.

Current valuations lack 1999 extremes

Leading tech companies trade at lower multiples than during the dot-com bubble—Microsoft peaked near 50x earnings in 1999 versus roughly 30x currently—while exhibiting fundamentally higher business quality and profitability.

Differentiated models beat trend exposure

Investors mistake riding secular growth waves for owning quality businesses, but durable competitive advantages and pricing power matter more than simply participating in technological shifts.

Strategic positioning varies by ecosystem role

Apple's licensing strategy may prove the 'safe path' compared to building proprietary LLMs, while Meta benefits from internal customer visibility, and some players may eventually question the strategic necessity of owning foundation models.