WTF (What's the Future) for AI - Insights and Action from Mary Meeker's AI Trends 2025 Report

The AI Trends 2025 report by Mary Meeker (https://www.bondcap.com/report/pdf/Trends_Artificial_Intelligence.pdf) paints a picture of an AI landscape evolving at unprecedented speed and scale. As AI Products move from experiments to mass‑market adoption, leaders must balance near‑term wins with mid‑ and long‑term bets.

The sections that follow:

1. Meta Insights & Patterns
2. Key Trends for AI Product Leaders
3. Futurecasting & “What’s the Future” (WTF)
4. Three Horizons Framework
5. Actionable Recommendations

Meta Insights & Overarching Patterns

“The pace and scope of change related to the artificial intelligence technology evolution is indeed unprecedented…often supported by user, usage, and spending charts that go up and to the right.”

– Mary Meeker, AI Trends 2025

From Meeker’s 340‑page compilation, three high‑level patterns stand out:

Compound Acceleration

• Training data for AI models has grown by ~260% annually over 15 years; compute used to train those models has expanded by ~360% per year over the same period; and algorithmic efficiency improvements account for ~200% compute gains per year. In other words, data + compute + better algorithms = compounded breakthroughs.
• As a historical reference, consider how global GDP took centuries to climb from $500 billion to $2 trillion but then exploded in the last two centuries with successive technology waves (printing press → steam engines → electrification → internet → cloud). AI is riding that curve and sharply bending it upwards.

Hyper‑Rapid User Adoption

• ChatGPT amassed 800 million weekly active users in just 17 months, compared with Google Search taking 11 years to reach similar annual query volumes. This reflects a user adoption curve ~10× steeper than previous major digital services.
• The 90% adoption mark for ChatGPT app happened in ~3 years, versus ~23 years for the internet to reach the same penetration outside North America.

Global, Fragmented Competitive Landscape

• Major tech incumbents (USA) and emerging players (China, open‑source communities) are racing to build the next generation of AI infrastructure—agentic interfaces, enterprise copilots, and autonomous systems.
• By early 2025, Chinese language models (e.g., Alibaba’s Qwen 2.5, ByteDance’s CodeFuse) matched Western LLMs on key benchmarks. At the same time, open‑source models (LLaMA, Stable Diffusion, etc.) proliferate, lowering barriers for startups.
• The upshot: proprietary MOPO (model‑operational‑platform‑offering) advantages are shrinking. Differentiation will increasingly rely on proprietary data, domain expertise, user experience, and ecosystem partnerships.

Together, these patterns imply that AI’s runway is steeper and more interconnected than any prior innovation wave.

The rest of this article unpacks what that means for AI Product strategy.

Key Trends for AI Product Leaders

Below are the top trends from Meeker’s report—with strategic context for leaders:

AI as a Ratchet for Scale

• Phenomenal User Growth: ChatGPT’s 800 million weekly users (April 2025) dwarfs any prior consumer app. Hyper‑scale is achievable fast.
• Strategic Implication: Build for scale from day one. Architect for millions of concurrent users, global localization, and high availability. If your AI feature or product goes viral, you must be ready.

From “AI as a Feature” to AI‑Native Platforms

• Platform Shift: Simply bolting on a chatbot to a legacy workflow is no longer sufficient. AI‑native products (e.g., GitHub Copilot, Notion AI, Canva Magic Write) reimagine a category around AI from the ground up.
• Strategic Implication: Ask yourself, “If I were starting from scratch today, how would I design my product with AI as a foundational pillar?” Those who answer this will define the next wave of platforms.

Open‑Source & Global Competition

• Feature‑Parity Pressure: Open‑source LLMs and Chinese models are fast‑catching up. Meeker notes that open‑source LLMs now rival some proprietary offerings on benchmarks.
• Strategic Implication: Leverage open‑source to accelerate development, but shift differentiation to unique data (e.g., vertical‑specific data), refined domain fine‑tuning, and bespoke user experiences (UX). Global competition means you must keep a finger on emerging ecosystems both in the West and in Asia.

Infrastructure & Cost Management

• Compute Costs: Training a frontier model in 2024 could cost $100 million–$1 billion. Inference cost (per token) fell ~99% in two years, but total compute demand is skyrocketing.
• Strategic Implication: Treat compute as a strategic asset—negotiate preferred cloud discounts or consider in‑house GPU clusters. Optimize models for cost (prune, distill, batch inference). Identify monetization paths (e.g., usage tiers, premium features) early—“growth at all costs” without a path to revenue invariably leads to unsustainable burn.

Human‑AI Workforce Evolution

• Job Transformation: From 2018 to 2025, AI‑related job postings in IT jumped +448%, while non‑AI IT roles declined 9%. Rather than wholesale replacement, many knowledge tasks (writing, coding, design) are being augmented or automated.
• Strategic Implication: Redefine your user personas: if your target user is a “content creator,” by 2026 they will expect AI to handle first drafts. Position your product to empower users, not compete against them. Internally, upskill talent toward “AI+X” roles (e.g., “prompt engineer,” “ML ops architect,” “AI ethicist”).

AI in the Physical World

• Embodied AI: Autonomous vehicles, warehouse robots, drone delivery—all moving from pilot to limited deployment. Meeker shows that autonomous taxi share in San Francisco reached ~8% by Q1 2025.
• Strategic Implication: If your product touches real‑world processes (logistics, manufacturing, healthcare), begin exploring AI + IoT / robotics integrations. Build partnerships with sensor or hardware vendors to co‑create. Even if you’re a digital‑only product, consider how “AI at the edge” (smart devices) could provide new data or use cases.

Ethics, Trust & Regulation

• Trust Gap: “Hallucinations,” biased outputs, data privacy breaches, and misinformation are top concerns. Regulatory frameworks are still forming—but will accelerate.
• Strategic Implication: Incorporate responsible AI practices early:

** Transparency (label AI outputs; provide confidence scores)

** Bias audits (test models on diverse data slices)

** Privacy safeguards (user consent, data minimization)

** Human‑in‑the‑loop oversight for high‑stakes decisions

Demonstrating robust governance is not only responsible but also a market differentiator—especially for enterprise and regulated industries.

Futurecasting: “What’s the Future” (WTF) & Long‑Horizon Signals

Meeker’s report offers concrete, data‑backed glimpses of AI’s capabilities in 2025–2035.

Below are key “WTF” observations:

By 2030 – AI‑First World

Meeker asked ChatGPT (May 2025) to list “Top Ten Things AI Will Likely Do in Five Years.” Highlights include:

1. Seamless Multimodal Interaction: Conversational interfaces that understand text, images, video, and audio together—your AI can analyze a photo while you ask questions about it in natural language.
2. First Drafts as Commodity: AI will generate everything from marketing copy to code boilerplate. Humans will pivot to curating, refining, and “meta‑thinking.”
3. Persistent AI Personas: AI agents that “remember” past interactions—no more resetting context each session. They’ll learn user preferences over time and proactively offer suggestions (e.g., “I see you often search for vegan recipes; here are three new ideas”).
4. Real‑Time Translation & Localization: Instantly translate spoken or written language with near‑human nuance, leveling the global creativity playing field.
5. Ambient Intelligence: AI running in the background of daily life—smart glasses that overlay context, or voice assistants that anticipate needs based on location/time without explicit prompting.

“We define the public launch of ChatGPT in November 2022 as AI’s ‘iPhone Moment’…AI is now a compounder on internet infrastructure.”– Meeker, p. 24

By 2035 – Prelude to AGI

Looking ten years out, Meeker’s team asked ChatGPT, “What Ten Things Will AI Likely Do in Ten Years?” The answers venture into AGI‑adjacent territory:

1. Autonomous Scientific Discovery: AI designing experiments, running millions of simulations, and proposing breakthrough hypotheses in drug discovery, materials science, and climate modeling—often faster than human researchers.
2. AI‑Designed AI: Early forms of autoML today will mature into AI that architects new AI architectures, optimizing compute and performance beyond human intuition.
3. Human‑Level Reasoning: AI that can reliably pass Turing‑style evaluations across a broad set of tasks—translating, summarizing, strategizing, and even engaging in abstract planning akin to an expert in multiple domains.
4. Full‑Context Personal Assistants: Your AI avatar knows your schedule, emotional state, and long‑term goals, offering personalized coaching (health, career, finance) around the clock.
5. Ethical & Governance Imperatives: As AI autonomy deepens, society will demand new legal/ethical guardrails—e.g., “AI Accountability Acts” requiring transparency on decision logic, or “Right to Human Review” in critical areas (healthcare, law, finance).

These are not certainties but signposts indicating where the research and investment are leading. Even if AGI (true general intelligence) remains elusive by 2035, incremental advances toward that capability will reshape entire industries.

AI Product leaders must ask: What happens to our business if an AI can outperform us on our core tasks? Preparing now—through strategy, partnerships, and ethical frameworks—will be critical.

Three Horizons Framework for AI Strategy

To coordinate effort across near, mid, and long timeframes, I adapted McKinsey’s Three Horizons model to the AI context. Below is a custom graphic summarizing how to align initiatives to each horizon.

Horizon 1 (2024–2025): Integrate & Optimize

Focus: Embed proven AI capabilities into current products; optimize for cost and performance; ensure reliable infrastructure.

Examples:

• Launch an AI chat assistant for customer support with fine‑tuning on proprietary transcripts.
• Replace rule‑based recommendations with a small, optimized recommendation model for in‑app suggestions.
• Implement usage‑based pricing or premium tiers to offset cloud compute costs.
• Establish a basic AI governance checklist (bias checks, user transparency).

Horizon 2 (2025–2030): Expand & Innovate

Focus: Build AI‑native products or new business lines; scale into enterprise/vertical solutions; deepen data moats.

Examples:

• Spin up a separate AI‑driven platform (e.g., an AI‑powered analytics suite) that can eventually replace legacy offerings.
• Partner with hardware/IoT vendors to integrate edge‑AI capabilities (e.g., AI inference on device).
• Acquire or license proprietary datasets to train vertical‑specific models (e.g., healthcare, legal).
• Launch a “Customer AI Academy” to train clients on advanced AI features (user enablement).
• Build an AI ethics center of excellence to lead on responsible AI in your industry.

Horizon 3 (2030–2035): Pioneer & Transform

Focus: Invest in aspirational R&D (moonshots) that could reshape the industry; build scenario plans for AGI‑adjacent breakthroughs; define long‑term trust and governance models.

Examples:

• Fund internal R&D exploring self‑improving AI systems or advanced multimodal reasoning prototypes.
• Set up strategic labs to investigate quantum computing for AI or neuromorphic hardware partners.
• Participate in multi‑company R&D consortia to shape AI regulation and standard‑setting.
• Develop internal talent pipelines (PhD fellowships, research residencies) to stay at the edge of AI science.
• Create “Ethics by Design” frameworks anticipating policy changes (e.g., “Right to Explanation,” AI safety audits).

By explicitly categorizing initiatives into these three horizons, AI Product leaders can balance short‑term product wins with mid‑term growth engines and long‑term transformation‑level bets. It prevents the common pitfall of being trapped in day‑to‑day firefighting (Horizon 1) and losing sight of the future, or vice versa.

Actionable Recommendations

Drawing on the above trends, Futurecasting, and Three Horizons, AI Product Leaders should consider the following strategic playbook:

Run Fast, Learn Faster (H1):

• Tactics: Launch AI features in controlled beta; measure usage, satisfaction, and ROI; optimize costs immediately.
• Goal: Achieve parity—or better—with competitor AI enhancements within 6–12 months.
• Key KPI: Time to roll out & iteration cycle (target < 2 weeks/feature update), user engagement uplift, cost per active user.

Build AI‑Native Next‑Gen Products (H2):

• Tactics: Dedicate a skunkworks team to explore AI‑first product ideas; partner with cloud/hardware vendors for preferential deals; acquire or co‑develop proprietary datasets.
• Goal: Launch one new AI‑native revenue stream by 2027 (could be a separate brand/business unit).
• Key KPI: Revenue from AI‑native product as % of total (> 15%), enterprise adoption rates, vertical customer wins.

Double Down on Data Moats (H2):

• Tactics: Integrate data‑collection hooks into every user flow (with clear consent); fine‑tune open models on your domain data; invest in data‑cleaning pipelines.
• Goal: Achieve a model performance delta (benchmarked on key tasks) that is > 10% better than generic, open‑source models by 2028.
• Key KPI: Model accuracy improvement on proprietary tasks, number of unique data sources in use, reduction in reliance on third‑party APIs.

Embed Responsible AI (H1→H2):

• Tactics: Create an “AI Ethics Playbook” (bias audit checklists, privacy guidelines, transparency docs); institute a small ethics review board.
• Goal: Achieve 100% transparency labeling on AI outputs; zero major ethical incidents; compliance with emerging regulations in key markets (EU AI Act, US guidelines).
• Key KPI: % of AI features passing bias audits, time to resolve user trust incidents, compliance score on external audits.

Cultivate Talent & Culture (H1→H2→H3):

• Tactics: Recruit top AI talent (incentivize via equity and autonomy); upskill existing employees with experiential “AI hackathons”; partner with universities for research fellowships.
• Goal: Grow your in‑house AI engineering team by 5× by 2027; publish 2–3 joint research papers by 2028.
• Key KPI: Number of AI engineers (year‑over‑year), average time to train a non‑AI engineer to proficiency, external R&D collaborations.

Invest in Moonshots & Monitor Signals (H3):

• Tactics: Allocate 5–10% of R&D budget to exploratory projects (e.g., autoML, quantum computing for AI); subscribe to leading AI research feeds; join policy roundtables.
• Goal: Have at least one prototype in areas like AI‑driven drug discovery or advanced reasoning by 2032.
• Key KPI: Number of Horizon 3 prototypes incubated, external citations or partnerships formed, key “leading indicators” (e.g., open‑source AGI projects launching).

Embrace Agile Partnerships & Ecosystems (H1→H2):

• Tactics: Offer APIs or SDKs for third‑party developers; join industry consortiums; co‑market with hardware/IoT vendors.
• Goal: Cultivate an ecosystem of 100+ apps/plugins using your AI platform by 2026; secure two strategic alliances by 2028.
• Key KPI: Number of ecosystem partners, revenue share from partner‑built solutions, API calls/transactions per month.

Stay Customer‑Centric (All Horizons):

• Tactics: Conduct frequent user research sessions (focus on how AI changes workflows); provide in‑product AI tutorials; maintain an “AI feedback channel” for real‑time improvement requests.
• Goal: Achieve > 80% user satisfaction for AI features; reduce customer support escalations related to AI outputs by 50% in one year.
• Key KPI: Customer Net Promoter Score (NPS) for AI features, AI feature usage retention rate, frequency of user‑reported AI issues.

Conclusion

Mary Meeker’s AI Trends 2025 report underscores a landscape where AI evolves faster than ever, adoption curves are steeper, and competition is global and multifaceted. From compounding compute/data growth to hyper‑rapid user adoption and emergence of AI‑native platforms, the signals are clear: the era of “AI as a sidecar” is ending; AI is becoming the engine of innovation across industries.

By organizing strategic efforts via a Three Horizons framework—emphasizing short‑term integration (Horizon 1), mid‑term expansion into AI‑native products (Horizon 2), and long‑term transformation bets (Horizon 3)—leaders can both capitalize on current momentum and prepare for tectonic shifts (e.g., the approach of AGI‑adjacent capabilities).

Ultimately, success in the next decade will hinge on the ability to:

• Move swiftly on near‑term AI enhancements while tightly managing costs.
• Invest in new AI‑native offerings and data moats that fuel sustainable, differentiated value.
• Embed Responsible AI practices to maintain trust as regulation catches up.
• Cultivate talent and partnerships to innovate and adapt as the technology evolves.
• Keep a vigilant eye on future signals and be ready to pivot if the horizon shifts.

The AI transformation is not a gradual wave but a rapid series of surges. Product and business leaders who heed Meeker’s data‑driven insights, align their strategy across horizons, and commit to building responsibly will be best poised to lead in 2025 and beyond. The future is coming faster than ever—start adapting now.w.