Is AI Still in 1995? Gene Munster and Doug Clinton on the Next Phase of the AI Boom

Episode Summary

Executive Summary: The discussion argues AI is still early but already transforming enterprise software, coding, and investing. The hosts frame AI as “electricity into intelligence,” expect a large but not yet peaked AI bubble, and emphasize that power, data centers, and energy infrastructure are the real bottlenecks. They also contend AI will disrupt knowledge work and reshape stock picking, with leading frontier models outperforming humans in specific financial tasks.

Main Topics: AI as an electricity-to-intelligence conversion layer (Priority: 5/5): The speakers frame AI as a fundamental industrial transformation where power is converted into intelligence, making energy and compute the core constraints on growth. AI adoption, coding agents, and enterprise utility (Priority: 5/5): They argue the breakthrough came when coding agents made AI useful enough for enterprises, pushing real productivity adoption and accelerating demand. Model competition and frontier capabilities (Priority: 4/5): The conversation compares GPT, Claude, Gemini, Grok, Meta, and Chinese/open-source models, noting that frontier models are converging but still differ by task. AI’s impact on jobs and knowledge workers (Priority: 5/5): The speakers expect more near-term disruption and unemployment among knowledge workers than prior tech cycles, but believe labor markets will adapt over time. AI investing: infrastructure, power, and hyperscaler capex (Priority: 5/5): They argue the best current opportunity is in power generation, nuclear, storage, and data center buildout, since AI demand is still outpacing infrastructure. Intelligent Alpha’s AI stock-picking benchmark (Priority: 4/5): The guests explain their system for testing 12 models on earnings direction and stock selection, showing that model quality varies by task and can be measured systematically. IPOs, market structure, and space-related opportunities (Priority: 3/5): They discuss how large AI IPOs could affect index flows and highlight speculative but potentially transformative space applications like orbital data centers and drug manufacturing.

Key Arguments: AI is effectively converting electricity into intelligence, so demand for power should scale with demand for intelligence. The AI cycle is still early—closer to 1995/1996 than 1998—and likely has several more years before a true bubble peak. Enterprise adoption accelerated after coding agents made AI materially useful; this is now a utility phase, not just experimentation. Frontier models are improving rapidly; today’s models are far more capable than a year ago and may soon perform like experienced professionals. There will be meaningful near-term disruption in knowledge-worker employment, especially for marginal performers who do not adopt AI. AI winners may not be zero-sum because capacity constraints mean even second-best models can monetize if they have available compute. The strongest investment theme is infrastructure: power generation, nuclear, storage, cooling, and data centers remain underbuilt relative to demand. Model performance differs by task: GPT is best for some benchmark predictions, Claude is strong for ideation/stock picking, and DeepSeek can outperform on certain earnings-driven portfolios. Large AI IPOs could create index rebalancing pressure and serve as a signal that private-market AI valuation and public-market appetite remain strong.

Data Points: AI trade cycle timing: 1995/1996 vs. 1998 - Doug says the AI market still feels closer to the early internet stage than a later-stage bubble peak. Knowledge-worker unemployment outlook: Next 5 years - He expects more acute knowledge-worker unemployment than occurred during mobile or the internet era. Enterprise revenue ramp for Anthropic: $9B to $45B run rate - Cited as evidence of rapid enterprise adoption of AI coding tools and models. Inference budget burn: First 4 months of the year - Uber and ServiceNow CTOs reportedly exhausted their annual inference budgets early because usage exceeded expectations. Models tested: 12 models - Intelligent Alpha’s earnings benchmark uses a dozen frontier and open-source models. Stocks tested: ~700 stocks - Their benchmark evaluates models across roughly 700 stocks each quarter. OpenAI model ranking: GPT 5.5 top - Doug says GPT 5.5 is currently the best model overall in their testing. Other model leaderboard: Opus 4.7, Gemini 3.1, Grok 4.3 tied-ish - The speakers describe the rest of the frontier pack as close behind GPT 5.5. Anthropic model leap: Opus 4.6/4.7/4.5 progression - They say each new model generation has been incrementally better than the last. AI task coverage: 95% to 98% - Doug claims current models can answer or figure out most tasks with decent accuracy. Capex expectation change: 10% to 70% growth - Hyperscaler capex growth expectations for calendar 2026 were revised upward dramatically. Another capex forecast: 10% to 20-30% growth - They expect next year’s capex growth to exceed Street expectations even if not as high as the prior year’s surge. U.S. energy output growth (1958-mid-1970s): ~7% annually - Used to show a historical period of rapid power expansion during the early nuclear era. Projected U.S. energy output growth: ~3% annually over 7-10 years - Cited from White House/Goldman research as a likely coming energy investment cycle. IPO reference: Cerebras stock up 108% intraday - Used as evidence that the market is receptive to high-profile AI and infrastructure IPOs. Aramco IPO reference: ~$1T+ market cap; stock up 30% initially - Historical comparison for what a mega-IPO could do to market structure.

Pivotal Quotes: "What AI really means, just as a first principle, is we're converting electricity into intelligence right now." — Doug: Used to frame AI as a physical infrastructure and energy story, not only a software story. "I do believe we're going to see in the near term, we'll call it a five years, more acute knowledge worker unemployment than we saw around mobile or the internet." — Gene: Their clearest statement on short-term labor disruption from AI. "If you go and use any of these models today, they are capable of probably answering or figuring out, you know, 95 to 98% of whatever you would throw at it with pretty decent accuracy." — Doug: Supports the claim that AI is already broadly useful enough to qualify as general intelligence-like in practice.

Implications: AI investment is shifting toward power, compute, and data-center infrastructure, while enterprise adoption and labor disruption accelerate. For investors and workers alike, the message is to follow the energy and model leaders, and adapt quickly or risk being left behind.