Merryn Talks Money• Apr 13, 2026• 37:32Interview

Why You Should Wait Out AI’s Super-Spending False Start | Merryn Talks Money

From Merryn Talks Money

Janusz Maretski•AI Partner, Aaron Innovation Capital

Executive Summary

The current AI paradigm of scaling large language models (LLMs) has hit fundamental limits due to a scarcity of high-quality training data, unsolvable issues like hallucination, and an inability to learn continuously.
The hundreds of billions of dollars being invested in data centers for this generation of AI represents a 'catastrophic misallocation of capital', with market signals like Oracle's CDS reflecting this risk.
Top AI researchers (LeCun, Silver, Sutskever) have left major tech companies to found startups focused on next-generation, brain-inspired AI that is vastly more power-efficient and capable of continual learning.
A new wave of AI is emerging from companies like Fractal Brain AI and Innate AI, which use novel architectures to solve the core problems of current models and are already outperforming them on benchmarks.

12 quotes

Concerns Raised

The current approach of scaling LLMs is a technological dead end with unsolvable flaws like hallucination and catastrophic forgetting.
Massive investment in data centers for current-gen AI is a 'catastrophic misallocation of capital' that will lead to significant financial losses.
The automation of entry-level software roles by AI could create a long-term skills gap for senior engineers.
Next-generation, continually adapting AI systems present new and 'spooky' safety risks as they cannot be permanently outsmarted.

Opportunities Identified

Investing in startups developing next-generation, brain-inspired AI architectures that can learn continuously.
A potential arbitrage strategy: shorting companies over-leveraged in data center expansion (e.g., Oracle) and going long on capital-light tech firms.
Investing in 'picks and shovels' companies that improve data center efficiency (cooling, algorithms) to profit from the current build-out.
The rise of powerful, efficient AI models that can run locally on laptops, creating new markets and disrupting the cloud-centric model.

Key Themes

The Limits of Scaling and Data Scarcity

The AI industry has exhausted the supply of high-quality, diverse training data, a problem that emerged as early as 2022 with the training of GPT-4. Training new models on recent, LLM-generated internet content leads to 'model collapse,' where performance degrades, indicating that simply scaling up current architectures is a dead end.

This challenges the prevailing industry strategy and suggests that future breakthroughs will not come from bigger models or more data, but from fundamentally new approaches.

Catastrophic Capital Misallocation in AI Infrastructure

The guest argues that the massive capital expenditure on data centers to support current-generation LLMs is a historic mistake. This is because the underlying technology is fundamentally flawed and is being superseded by more efficient models that can run locally on laptops.

Investors should be wary of companies with heavy exposure to data center build-outs for AI, as this infrastructure may become obsolete, posing a significant financial risk.

The AI Research Exodus and Paradigm Shift

Leading AI researchers are 'jumping ship' from major tech firms like Meta and DeepMind to create startups focused on solving AI's core challenges. This shift from an 'age of scaling' back to an 'age of research' signals a major pivot in the industry towards new, brain-inspired architectures that prioritize continual learning and efficiency.

The intellectual and innovative center of AI is moving away from established giants towards a new ecosystem of startups, which is where the next major breakthroughs are likely to originate.

Emergence of Next-Generation, Continually Learning AI

A new class of AI systems is being developed that mimics biological processes, such as Hebbian learning and neural growth, to enable continual learning without 'catastrophic forgetting'. These models are 3-4 orders of magnitude more power- and data-efficient than current LLMs and are already demonstrating superior performance.

This next generation of AI promises to solve the key limitations of today's models, unlocking new capabilities while also disrupting the business models built around massive, centralized compute.

Get started free

Topics

AI Investment Strategy Large Language Models (LLMs)Continual Learning Hallucinations (AI)Catastrophic Forgetting Model Collapse Data Scarcity (AI)Data Center Economics Capital Misallocation Next-Generation AI Brain-Inspired AI Neural Networks Backpropagation Hebbian Learning AI Efficiency

Processed Apr 13, 2026 yt-dlp + mlx-whisper + Gemini