Why $3T in AI infrastructure spending might miss the mark

Newsweek’s AI Impact newsletter explores whether today’s AI hyperscalers—OpenAI, Google, Microsoft, Meta, and Anthropic—may be making the same mistake as early automotive pioneers who envisioned centrally-owned vehicle fleets rather than individual ownership. The analysis questions whether the current model of massive data centers serving cloud-based AI will prevail, or if localized AI alternatives could emerge as the dominant paradigm, similar to computing’s evolution from mainframes to personal devices.

The big picture: AI companies are betting trillions on centralized infrastructure, but history suggests transformative technologies often evolve in unexpected directions.

• OpenAI CEO Sam Altman projects “trillions of dollars over time” in AI infrastructure spending, while Morgan Stanley predicts global data center investment will approach $3 trillion between 2025 and 2029.
• Early automotive pioneers like Thomas Edison correctly predicted cars would replace horses but missed how individual ownership would reshape society through suburban sprawl.
• The Electric Vehicle Company operated over 1,000 electric cabs from central hubs in 1897, mirroring the horse-and-carriage rental model rather than anticipating personal car ownership.

Why this matters: The current AI boom channels massive capital into fast-evolving technology deployed on infrastructure with 12-18-month depreciation cycles, raising fundamental questions about the industry’s direction.

• Small language models offering economical, specialized alternatives to ChatGPT represent a “quieter revolution” in localized AI.
• This parallels computing’s evolution from shared institutional mainframes to personal desktops to smartphones.
• The most important unanswered question may be whether AI’s greatest impact comes from frontier cloud models or “good enough” agents in personal devices.

Research breakthrough: University of Pennsylvania psychology professor Philip Tetlock’s forecasting research reveals how AI can enhance human prediction accuracy.

• Human forecasters working with large language model “advisors” improved their accuracy by up to 41 percent in tests.
• “The clear conclusion was that just the activity of interacting with a state-of-the-art LLM was helpful to all human participants under all circumstances,” Tetlock explained.
• He predicts with “80 percent confidence” that LLMs themselves could become superforecasters within five years.

Real-world applications: Practical AI use cases demonstrate the technology’s current capabilities and limitations.

• Palantir deployment strategist Hind Kraytem used AI to generate culturally appropriate Japanese business emails by providing tone-rich bullet points rather than literal translations.
• Hospital for Special Surgery’s Chief Digital Officer Ashis Barad successfully used ChatGPT’s agent mode to research restaurants, check availability, and make reservations for a family gathering, though the process wasn’t perfect.

Industry developments: Several trends indicate both AI adoption challenges and growing skepticism.

• Many organizations struggle to move beyond pilot projects due to culture, governance, and leadership alignment issues rather than technical limitations.
• Automation is displacing early-career jobs as companies use AI for data analysis, scheduling, and communications previously handled by entry-level staff.
• Public sentiment toward generative AI is declining, with 43 percent of Americans now saying AI is more likely to harm than help, as creators push back against uncredited data use.