Divya van Mahajan

The State of AI - Mid-2025

ai-daily-brief-podcast

State of AI Mid-2025

Overview

This episode of the AI Daily Brief podcast provides a structured mid-year survey of where artificial intelligence stands as of mid-2025, drawing on two major reports: a macro investment analysis from venture/investment firm Koatu (described as a Mary Meeker-style presentation) and Menlo Ventures’ State of Consumer AI 2025 report. The host synthesises private market trends, public market trends, and consumer adoption data to argue that AI is in the midst of a sustained “super cycle” with no signs of slowing. No individual speaker name or institutional affiliation beyond the show itself (AI Daily Brief) is provided.

Source URL: (not provided)

Prerequisites

  • Basic familiarity with venture capital and public equity markets (IPOs, M&A, SPACs, market capitalisation)
  • Understanding of key AI companies: OpenAI, Anthropic, Google (Gemini), Microsoft (Copilot/Azure AI), Meta, xAI
  • Awareness of major AI product categories: large language models (LLMs), reasoning models, coding assistants, image generation
  • Familiarity with terms such as “tokens,” “agentic AI,” and “vibe coding”
  • General knowledge of macroeconomic concepts: interest rates, debt-to-GDP ratio, inflation, capital expenditure (CapEx)
  • Awareness of recent AI industry milestones: DeepSeek release (January 2025), ChatGPT Deep Research launch, Goldman Sachs AI ROI report (July 2024)

Main Points

1. Private Markets Are Overwhelmingly Dominated by AI

  • AI now accounts for over 50% of all private market funding in 2025 to date.
  • The AI sector is emerging from an “exit desert” caused by post-ZIRP conditions; early signs of an IPO window, increased M&A, and the return of de-SPACs signal improving liquidity.
  • The top 10 private AI companies grew in combined valuation by 130% (from $283 billion to $658 billion) in the year ending June 2025.
  • $10 billion funding rounds have become normalised for new model-layer startups.
  • Time to $10 million in revenue has compressed from 10 years to 12 months.

2. Anthropic as a Case Study in Hypergrowth

  • Anthropic reached $1 billion in annualised revenue approximately 21 months after founding.
  • Subsequent billion-dollar revenue milestones arrived in 3 months, 2 months, and then 1 month respectively.
  • As of the report’s data, Anthropic is at $4 billion annualised revenue and accelerating, driven significantly by demand for AI coding tools built on its Claude models.

3. Public Markets: The Action Is Elsewhere, but Subsectors Outperform

  • The top 10 public AI companies grew combined market cap by only 10% (from $18 to $20 trillion) over the same period — compared to 130% for their private counterparts.
  • The Magnificent Seven as a group is down 1% year-to-date; four of the seven are negative.
  • Three outperforming public AI subsectors identified by Koatu:
    • AI Semiconductors (e.g., Broadcom, TSMC): +12%, driven by rising token demand
    • AI Software (e.g., Palantir, Oracle, Snowflake): +17%, driven by agentic AI workflows enabled by reasoning models
    • AI Power/Energy (e.g., GE Vernova, Constellation, Vistra): +18%, driven by electricity demand and hyperscaler contracts

4. The “Age of Reasoning” — Token Consumption Inflection

  • Koatu characterises the current period as the “Age of Reasoning,” marked by a sharp increase in token consumption following the release of reasoning models in Q4 2024.
  • Microsoft processed over 100 trillion tokens in Q1 2025 (per Satya Nadella’s April 30 earnings call), a 5× year-over-year increase; 50 trillion tokens were processed in March 2025 alone.
  • The share of U.S. businesses with paid AI subscriptions jumped from ~25% in Q4 2024 to 42% in Q1 2025 (per Ramp spending data), coinciding with the emergence of reasoning models.
  • ChatGPT’s growth surged following the launch of its Deep Research feature and updated image generation.

5. Macroeconomic Risks and Tailwinds

  • Three identified risks to the AI market cycle: overvaluation, tariff-driven erosion of U.S. exceptionalism, and ballooning government deficits.
  • Key tailwinds: elimination of the AI diffusion export rule (replacing tight chip export controls with broader ally access), resulting in $250 billion–$1 trillion in Gulf state AI investment commitments; a nuclear and energy renaissance replacing policy paralysis.
  • Koatu’s structural transformation thesis:
    • Autos → Semiconductors
    • Oil → Electricity
    • Manufacturing factories → AI factories
    • Industrial revolution leadership → AI revolution leadership

6. The AI Productivity Flywheel Thesis

  • Koatu argues AI could trigger a virtuous economic cycle: AI productivity gains → lower unit labour costs → lower inflation → lower interest rates → higher GDP growth → higher tax receipts → lower debt-to-GDP ratio.
  • The Koatu Fantastic 40 slide identifies companies projected to lead public markets by 2030; OpenAI is listed at #9 with a projected $1.6 trillion market cap, and xAI at #35.

7. Consumer AI Adoption: Scale and Demographics

  • 61% of American adults have used AI in the past six months; 20% use it daily (Menlo, n=5,000 U.S. adults).
  • Menlo characterises this as “habit formation at an unprecedented scale,” no longer mere experimentation.
  • Generational breakdown: Gen Z leads overall adoption; Millennials are the heaviest daily users; nearly half of Baby Boomers have used AI, with 11% using it daily.
  • Employment and income correlation: 75% of employed adults use AI vs. 50% of unemployed; 74% of households earning $100K+ use AI vs. 53% of households under $50K.
  • 85% of college-age students report using AI.

8. Parents as an Unexpected Power-User Segment

  • 79% of parents have used AI, vs. 54% of non-parents.
  • 29% of parents use AI daily — nearly twice the rate of non-parents.
  • 34% use it for child care management.
  • Narrative gateway experiences (e.g., AI storytelling, image generation with children) are identified as a driver of broader personal adoption.

9. Tool Landscape: Convenience and First-Mover Advantage

  • ChatGPT leads general AI assistant usage at 28% of adult users.
  • Other tools by share: Gemini 23%, Meta AI 18%, Alexa 18%, Siri 16%.
  • Distribution reflects name recognition and pre-existing platform presence rather than capability differentiation.

10. What People Use AI For

  • Top use cases (each representing 14–19% of U.S. adults): writing emails, researching topics, managing to-do lists, general writing support, meal planning, managing expenses, note-taking, creating images, researching purchases, researching health questions.
  • AI penetrates most deeply into activities people were already doing.
  • Vibe coding is the standout example of AI enabling new behaviour: 47% of respondents use AI for coding for work/school; 41% for personal coding projects.

11. Barriers to Adoption Among Non-Users

  • Among the 39% of Americans not yet using AI:
    • 80% prefer people over AI
    • 71% cite data privacy concerns
    • 63% see no need
    • 58% don’t trust AI outputs
    • 48% don’t know how to use it
    • 40% perceive bias
    • 27% cite lack of access

12. Predictions for the Next Wave

  • Shift from generalised tools (ChatGPT-style) to specialised vertical tools.
  • Movement from AI assistants to full automation and agentic workflows, mirroring the enterprise trajectory.
  • Emergence of multiplayer/social AI experiences.
  • Significant growth of voice AI interfaces.
  • Physical AI entering the home.
  • Revenue model diversification beyond subscriptions.

Key Concepts

  • Age of Reasoning: Koatu’s term for the post-Q4 2024 period marked by the widespread deployment of reasoning models and the resulting surge in token consumption.
  • Reasoning models: AI models (e.g., OpenAI o-series, Claude 3.x) that perform extended internal “chain-of-thought” processing before producing outputs, consuming significantly more tokens per query.
  • Vibe coding: AI-assisted software development in which users describe intent in natural language and the AI generates functional code; identified as a key inflection point in both market revenue and consumer behaviour.
  • AI diffusion rule: A U.S. government policy controlling the export of advanced AI chips; its relaxation in 2025 expanded chip access to allied nations, enabling large foreign AI investment commitments.
  • Magnificent Seven (Mag7): The cohort of the largest U.S. technology public companies (Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, Tesla); used here as a benchmark against which AI subsector performance is compared.
  • AI super cycle: The thesis that current AI investment, adoption, and revenue growth represents a sustained, self-reinforcing economic expansion rather than a speculative bubble.
  • Productivity flywheel: Koatu’s model in which AI-driven labour productivity gains cascade into macroeconomic improvements including lower inflation, lower interest rates, GDP growth, and reduced government debt burden.
  • Koatu Fantastic 40: Koatu’s proprietary list of 40 companies projected to be among the largest public market companies by 2030.
  • ZIRP (Zero Interest Rate Policy): The near-zero interest rate environment that prevailed during the COVID era (2020–2021), which inflated venture valuations and exit volumes; its end created the subsequent “exit desert.”
  • De-SPAC: A method of taking a private company public by merging it with a Special Purpose Acquisition Company (SPAC) already listed on a stock exchange, identified as re-emerging as a viable path to liquidity.
  • Agentic AI / AI agents: AI systems capable of autonomously executing multi-step tasks with minimal human intervention, increasingly enabled by reasoning model capabilities.

Summary

At the midpoint of 2025, the central narrative across both private and public AI markets and consumer adoption is one of accelerating momentum rather than consolidation or correction. Private markets are capturing the majority of AI’s financial upside, with top private AI companies growing in value at 13 times the rate of their public counterparts; funding rounds of $10 billion have become routine, and hypergrowth trajectories like Anthropic’s are compressing traditional revenue timelines to a fraction of historical norms. In public markets, the legacy Magnificent Seven are underperforming while specialised AI subsectors — semiconductors, software, and energy infrastructure — are posting solid gains, driven by the structural shift Koatu labels the “Age of Reasoning,” in which reasoning model deployment has caused token consumption to surge 5× year-over-year and enterprise AI subscription penetration to jump sharply in a single quarter. On the consumer side, Menlo’s data confirms that AI has crossed from experimentation into habitual use for the majority of American adults, with parents, millennials, and employed higher-income individuals emerging as the most intensive users, and vibe coding representing the clearest example of AI creating entirely new behaviours rather than simply augmenting existing ones. The remaining barrier to universal adoption is primarily attitudinal and trust-based rather than access-based. Both reports converge on the same conclusion: the AI cycle is not slowing, the capital is not retreating, and the structural transformation of the economy — from energy infrastructure to software to government fiscal dynamics — is still in its early stages.