The central and most important competition in AI is the hardware arms race between NVIDIA's GPUs and Google's TPUs, which are currently the only viable alternative for large-scale training.
A crossover investing strategy that spans both public and private markets is essential in the AI era due to the unique mix of public and private competitors at every layer of the tech stack.
The AI buildout is not a 2000-style bubble because it is constrained by physical limits on energy and semiconductor supply, and the massive capital expenditures are already generating higher returns on invested capital for the spenders.
Traditional SaaS companies face an existential threat and must fundamentally change their business models to accept the lower gross margins inherent in compute-intensive AI products to survive.
AI scaling laws for pre-training remain intact, as proven by models like Gemini 3, and recent advances in reasoning have bridged an 18-month innovation gap, ensuring continued progress.
1996
Performs an early analysis of Cisco, which he notes increased twenty-fold in the subsequent four years, marking an early point in his tech analysis career.
Pre-2019 (Fidelity Tenure)
As a portfolio manager at Fidelity, demonstrates his crossover investing approach by leading three consecutive funding rounds in the private company Roku, despite market concerns about competition from tech giants.
2019
Founds Atreides Management, a firm managing $7 billion across public, private, and crossover strategies, institutionalizing his investment philosophy.
2022-Present
Public commentary becomes heavily focused on the AI technology stack, comparing the emergence of ChatGPT to Netscape Navigator and framing it as the beginning of a multi-decade investment cycle.
2024
Articulates a highly detailed thesis on the AI hardware market, identifying Google's TPU as NVIDIA's main competitor, detailing the physical constraints of the Blackwell transition, and making specific predictions about the future of AI models and infrastructure.
▶The AI Hardware Arms Race
Baker's analysis is intensely focused on the foundational layer of AI competition: the hardware. He frames the market as a primary battle between NVIDIA's GPUs and Google's TPUs, with both companies accelerating their roadmaps to an annual release cycle to outpace the development of custom ASICs by competitors.
This focus suggests that the primary value capture and most durable competitive moats in the current AI wave are being built at the silicon level, making analysis of chip architecture, manufacturing capacity, and data center infrastructure paramount for investors.
▶The Crossover Investing ImperativeApr 2026
Baker argues that the AI industry is unique because public and private companies are direct competitors at every level of the technology stack, from frontier models (OpenAI vs. Google) to infrastructure. He believes this dynamic makes a crossover investing strategy, which spans both public and private markets, not just relevant but 'paramount' to truly understand the landscape.
This implies that traditional public-only or private-only investment funds may have a critical blind spot, unable to fully grasp the competitive dynamics and technological shifts occurring across the public/private divide in AI.
▶AI's Economic Transformation and Creative Destruction
Baker posits that AI forces new, and often painful, economic realities on existing businesses. He repeatedly warns that traditional SaaS companies must accept lower gross margins to survive, while noting that the large companies investing heavily in AI are already seeing significant increases in their return on invested capital.
Baker's analysis points to a period of significant turmoil for incumbent software companies, where failure to adapt to AI-native, agent-based models could be an existential threat, analogous to brick-and-mortar retailers who failed to embrace e-commerce.
▶Physical World Constraints on a Digital RevolutionApr 2026
A core tenet of Baker's thesis is that the AI buildout is fundamentally limited by real-world resources like electricity, data center rack space and weight limits, and semiconductor manufacturing capacity. He argues these bottlenecks will moderate the investment cycle and may prevent the kind of speculative oversupply seen in the dot-com era's fiber optic boom.
This perspective provides a contrarian check on simplistic 'bubble' narratives, suggesting that the AI investment cycle's trajectory will be dictated as much by engineering and energy constraints as by market sentiment.