The Download: tracing AI-fueled delusions, and OpenAI admits Microsoft risks

2026-03-24 · Source: MIT Technology Review · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Software Development & Engineering · Depth: Intermediate, extended

Summary

This intelligence brief, dated March 24, 2026, covers several key technological developments and industry insights. Stanford researchers have analyzed chatbot user transcripts, suggesting AI can amplify delusion-like thoughts into obsessions, though the causal link remains unclear. OpenAI has acknowledged its reliance on Microsoft as a business risk in pre-IPO documents, while also pursuing automated research and challenging Google's search dominance. The US has banned new foreign-made consumer routers due to national security concerns. NVIDIA CEO Jensen Huang discusses the company's shift from GPU design to "AI factory" rack-scale co-design, emphasizing the importance of extreme co-design for scaling AI and addressing power consumption through efficiency and grid optimization. Huang also highlights the critical role of CUDA's install base and ecosystem as NVIDIA's primary competitive advantage, and his vision for AI agents as the "iPhone of tokens."

Key takeaway

For CTOs and VPs of Engineering planning AI infrastructure, recognize that NVIDIA's "AI factory" approach and CUDA ecosystem are defining the future of compute. Your strategy should account for the increasing complexity of rack-scale systems and the need for extreme energy efficiency. Proactively engage with utility providers to explore flexible power contracts that leverage off-peak grid capacity, and invest in systems capable of gracefully degrading performance during power constraints to ensure operational resilience.

Key insights

AI's rapid evolution necessitates extreme co-design, strategic ecosystem development, and proactive infrastructure planning to overcome scaling challenges.

Principles

• Install base defines an architecture's success.
• Intelligence scales primarily with compute.
• Simplicity is achieved by necessary complexity, not gratuitous additions.

Method

NVIDIA employs extreme co-design across the entire software and hardware stack, from chips to data centers, anticipating future AI model architectures and fostering a broad ecosystem through strategic partnerships and open-source contributions.

In practice

• Consider AI's potential to amplify user delusions.
• Prioritize energy efficiency in AI infrastructure design.
• Explore dynamic power allocation for data centers to utilize excess grid capacity.

Topics

• AI Ethics
• AI Infrastructure
• AI Scaling Laws
• Semiconductor Manufacturing
• AI Agents

Best for: Investor, CTO, VP of Engineering/Data, Executive, AI Architect, Tech Journalist

Related on AIssential

• GPT: The “wheels are coming off” not because AI is useless, but because the current AI buildout assumes that technical scale, capital expenditure and vendor optimism can outrun every other constraint. — The AI economy faces critical constraints in chips, energy, data, and governance, shifting focus from model scale to infrastructure reality.
• Bloomberg Tech at CES — AI compute demand is rapidly accelerating across cloud, enterprise, and consumer sectors, driving innovation in chip design and autonomous systems.
• NVIDIA CEO: AGI Is Already Here — NVIDIA's CEO declares AGI is here, emphasizing extreme co-design and CUDA's foundational role in scaling AI capabilities.
• NVIDIA GTC Studio with Insights from Schneider Electric — AI can both consume and optimize energy, creating a circular relationship for sustainable scaling of AI infrastructure.
• The Hidden Challenges of Running AI at Scale in Production — Specialized AI clouds and infrastructure optimization are crucial for scaling AI from pilots to production workloads.
• AI Weekly Issue #476: Weekly Intelligence Briefing: Tech, AI & Policy — NVIDIA's success stems from extreme co-design, a vast CUDA install base, and anticipating AI's future needs across hardware and software.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT Technology Review.