No AI bubble: The real money starts in 2026’s inference phase

AI Investment Landscape: A Deep Dive into Growth, Infrastructure, and Future Potential

Key Concepts:

• AI Bubble: The concern that the rapid growth in AI-related stock valuations is unsustainable and will lead to a market correction.
• R&D Phase: The current stage of AI development focused on research and development, characterized by significant investment and limited revenue.
• Inference Stage: The upcoming phase where AI models are deployed in applications and generate revenue through usage.
• Large Language Models (LLMs): AI models capable of understanding and generating human language, like those developed by OpenAI and Google.
• General Purpose Computing (GPC): Computing infrastructure designed for a wide range of tasks, as opposed to specialized applications.
• TPU (Tensor Processing Unit): Google’s custom-designed AI accelerator chip.
• Behind-the-Meter Power: Power generation and storage solutions located on the customer’s side of the electricity meter, reducing reliance on the grid.
• Training Phase: The initial stage of AI development where models are created and refined using large datasets.

I. Debunking the AI Bubble Narrative

The discussion began by addressing widespread concerns about an “AI bubble.” Both Ivana Dylewska (Way In Spear Invest) and Beth Kindig (I.O Fund) strongly refuted this notion. Dylewska argued that current market multiples are actually lower than historical averages, contrasting this with typical bubble behavior where valuations expand ahead of fundamentals. She highlighted that, unlike past bubbles, Price-to-Earnings (P/E) ratios have been decreasing over the last several years. Kindig echoed this sentiment, emphasizing that the current phase is one of substantial Research and Development (R&D) investment by Big Tech companies, totaling hundreds of billions of dollars. This investment is not indicative of a bubble, but rather a necessary precursor to the upcoming “inference and monetization phase,” expected to begin in 2026. The comparison to the dot-com bubble of 2000 was explicitly dismissed.

II. The Shifting Dynamics Between Google and OpenAI

A recent debate centered on the perceived fortunes of Google and OpenAI. Initially, there was pessimism surrounding Google’s ability to compete in the AI space. However, by the end of the year, Google had become a market favorite. Kindig countered this binary view, asserting that both Google and OpenAI are poised to be winners, particularly during the inference stage. She explained that both companies are focused on embedding their Large Language Models (LLMs) within a wide range of applications, creating a proliferation of opportunities for revenue generation.

III. Financial Implications and OpenAI’s Rapid Monetization

The conversation then turned to the financial aspects of OpenAI, which recently raised $1.5 trillion. Concerns were raised about the pressure to deploy capital and the potential impact on companies like Oracle, which may be awaiting funding. Kindig countered these concerns, pointing to OpenAI’s unprecedented speed of monetization. She compared it to the rapid growth of mobile and cloud computing, stating that no other company has monetized as quickly. OpenAI is projecting hundreds of billions in revenue by the end of the decade.

IV. The Future of AI Hardware: NVIDIA and Chip Specialization

Dylewska identified NVIDIA as a continuing winner, maintaining its leadership position in General Purpose Computing (GPC). However, she predicted a “bifurcation” in chip development, with increasing specialization based on specific use cases. For example, Google’s Tensor Processing Units (TPUs) may be ideal for advertising applications, while NVIDIA chips remain better suited for more general-purpose AI tasks like those undertaken by OpenAI. This specialization will lead to a diverse landscape of chips optimized for different AI workloads.

V. The Critical Role of Power Infrastructure

The discussion shifted to the crucial issue of power supply for AI infrastructure. Initially, concerns revolved around potential power shortages. However, the focus has now shifted to leveraging existing power infrastructure and securing new capacity. Kindig highlighted the potential of Bitcoin miners, who have secured significant power capacity but haven’t yet fully contracted it out, as a valuable resource. She specifically mentioned Apply Digital, Terra Wolfe, and Iron as companies positioned to benefit from this trend.

A key example was Bloom Energy, which was previously featured on 60 Minutes. While the stock initially underperformed, it has recently experienced significant growth due to its ability to provide on-site power solutions, particularly solid oxide fuel cells that don’t rely on the traditional grid. Bloom’s ability to energize data centers for Oracle in just 55 days was cited as a testament to its capabilities. Kindig emphasized that AI is an “energy race,” and Bloom’s unique approach to power delivery gives it a competitive advantage.

VI. Understanding Inference: The Path to Monetization

The concept of “inference” was explained as the stage where AI models are deployed in applications and generate revenue. Dylewska clarified that the current phase is the “training phase,” which is compute-intensive and not directly monetizable. Inference, on the other hand, represents the opportunity to finally monetize these models through application usage. The speakers emphasized that the industry is on the cusp of this transition, with numerous potential winners emerging.

VII. Beyond Picks and Shovels: The Expanding AI Ecosystem

The conversation concluded with a brief mention of Rocket Labs, highlighting the continued growth potential in the space sector. This underscored the broader point that the AI revolution extends beyond the core technology companies and into adjacent industries.

Notable Quotes:

• Beth Kindig: “No other company has monetized as quickly as OpenAI.”
• Ivana Dylewska: “Multiples where we’re seeing them today, they’re actually quite muted and trading at the lower end of the historic range.”
• Beth Kindig: “AI is an energy race.”

Conclusion:

The discussion presented a bullish outlook on the future of AI, dismissing the “bubble” narrative and emphasizing the significant investment and innovation driving the industry forward. The key takeaway is that the current R&D phase is a necessary precursor to the highly lucrative inference stage, expected to begin in 2026. Successful investment strategies will focus on companies positioned to capitalize on this transition, including hardware providers like NVIDIA, power infrastructure companies like Bloom Energy and Apply Digital, and the AI model developers themselves, Google and OpenAI. The importance of securing adequate power infrastructure was repeatedly stressed as a critical factor for future growth.