Multi-Trillion Dollar AI Blueprint
Inside the AI Ascent 2025 Sequoia's Keynote
Alright, friends, buckle up.
When Sequoia Capital talks, the tech world tends to listen.
At their recent AI Ascent 2025 keynote, Pat Grady, Sonya Huang, and Konstantine Buhler didn’t just talk—they laid out a vision for a mind-bogglingly huge AI-driven future.
Forget your previous TAM estimations; we’re entering a new stratosphere.
I’ve distilled the core insights and my thoughts for you
The Scale Is Unprecedented. Rails Are Laid 🏔️
Pat kicked things off by essentially saying: hold onto your hats, because the AI tsunami is even bigger than we thought.
1. So What?
Remember the Cloud transition? It was big. When Cloud started, the software market was about $350 billion; Cloud itself is a $400 billion behemoth today.
AI is attacking both the $10 trillion global services market (where AI is currently a tiny $3 billion sliver) and the $650 billion software market (where AI is a $15 billion newcomer).
The starting point for AI’s disruption is an order of magnitude larger than Cloud’s. The endpoint in 10–20 years?
Absolutely massive.
2. Why Now?”
Physics of Distribution Have Changed.
Tech waves are additive: Semiconductors → Systems → Networks → Internet → Apps → Mobile → AI.
The ingredients are here: compute, networks, data, distribution, talent.
Crucially, the physics of distribution has transformed:
Awareness:
Cloud (Salesforce’s “No Software” logo) needed guerrilla marketing.
Mobile had the iPhone launch.
ChatGPT – the entire world knew instantly.
Desire (MAUs of discovery platforms like Reddit/X):
Cloud era – 0;
Mobile era – 4 million;
AI era – 1.8 billion.
Action (connected users):
Cloud era – 200 million.
Mobile – 1.4 billion.
AI era – 5.6 billion.
The rails are in place. And when the starting gun went off, there were no barriers to adoption…
The physics have changed.
3. “What Now?”
Race to the Application Layer 🏁
Yes, there’s white space, but foundation models are coming for it (tech-out approach).
Startups: go customer-back. Solve specific vertical or functional problems, tackle complexity, and leverage human-in-the-loop.
The Leone Merchandising Cycle (build moats!): Vision → Product → Engineering → Marketing → Sales → Support. It’s the whole business.
Metrics that matter:
Revenue: Beware vibe revenue. Is it real, sticky, creating durable behaviour change?
Margins: Focus on the slope, not the intercept. COGS will likely fall, but capture value by moving up the value chain.
Data flywheel: What business metric does it move? If you can’t answer, it’s BS.
“Nature hates a vacuum… You are in a run-like-heck business right now.
Now is the time to go at maximum velocity, all of the time”. Pat Grady
Adoption ✨
Sonya Huang reminds us about The “Her” Moment & the Abundance Era. She provided a reality check on current AI adoption and a glimpse of what is becoming possible.
Year in Review (Customer-Back)
AI-app engagement (DAU/MAU) is soaring. ChatGPT is nearing Reddit levels – a massive shift from last year’s “hype exceeds reality.”
We’re moving beyond viral fun (Ghibli-fying everything, anyone?) to real applications in advertising, education (visualising complex concepts instantly) and healthcare (hello, OpenEvidence).
Voice AI had its “Her” moment (see Sesame’s demo). The uncanny valley is being crossed, and it’s happening fast.
Year in Review (Technology-Out)
Pre-training is hitting a wall – the easy gains are gone.
New vectors for progress: reasoning models, synthetic data, tool use and agentic scaffolding.
Innovation is happening at that blurry product-model boundary (think OpenAI’s Deep Research and Google’s NotebookLM).
AI’s Killer Apps & the Abundance Era
The first cohort of AI killer apps has emerged (ChatGPT, Harvey, Glean, Sierra, Cursor, Abridge). The next wave is rising.
Prediction for 2025: Vertical agents will deliver deeper tech and customer value (for example, Xbow in cybersecurity, Traversal in DevOps, Meter in networking).
We’re entering an Abundance Era: As labour (human and AI-driven) becomes cheaper and plentiful, what becomes scarce? Taste – the ability to discern quality and craft truly valuable experiences.
Konstantine Buhler: Welcome to the Agent Economy 🤖🤝📈
Konstantine took us further into the future – it’s all about intelligent agents.
From Agent Swarms to an Agent Economy
Last year, the buzz was AI agents as individual machine assistants.
Now: “agent swarms” – multiple agents collaborating, sometimes even competing, to achieve complex goals.
Next leap: the Agent Economy. Agents will:
Transfer resources
Make transactions
Track reliability
Establish trust
How We Get There – Critical Tech Challenges
Persistent identity for agents and humans.
Seamless communication protocols – think TCP/IP for agents.
Secure trust: if you can’t meet an agent “face-to-face,” you need new trust mechanisms.
Living in the Agent Economy – Mind-set Shifts
Stochastic mind-set: manage a distribution of outcomes rather than a single answer.
Management as a computer skill: humans will orchestrate agent swarms.
Leverage over uncertainty: more leverage, less certainty.
Thoughts
The first thing that hits me is scale. AI is not just the next software platform; it is also coming for the ten‑trillion‑dollar services market. Cloud’s starting line looks almost quaint by comparison.
Second, the battlefield will be the application layer unless you are building foundation models. Deep domain expertise, workflow integration, and tight feedback loops will matter more than marginal gains in model quality.
Third, moats are holistic. Pat’s Leone Merchandising Cycle reminds me that code, data, brand, sales motion, and customer success compound together. Ignore any one of them, and the flywheel stalls.
Fourth, metrics must be real. Demos and vibe revenue are fun until the funding climate turns. If usage does not translate into revenue, margin expansion, or a measurable lift in a customer KPI, it will not last.
Fifth, speed is life. The vacuum will fill; the question is whether you fill it or let your faster competitor do so.
Finally, agents are inevitable. The winners will combine technical leverage with human taste – the ability to decide what should be built, not just what can be automated.
This AI Ascent felt less like a forecast and more like a field guide to the very near future. Sequoia’s message is crystal clear: the revolution is accelerating, the opportunities are vast, and the game is changing fast.
Time to build.
Post-credit Scene
If you’re still hungry for more, here’s what I have on my radar this week:
Sequoia Keynote i’m talking about
📺 Film
Now, truly valuable something for you, not just endless content, books, and podcasts.
NVIDIA just released a set of free online AI courses> Grab them while they’re hot:
Generative AI Explained – define generative AI, how it works, key applications, challenges and opportunities.
AI for All: From Basics to GenAI Practice – AI impacts industries like healthcare and autonomous vehicles, from machine learning to generative AI, creating music, images and videos.
Getting Started with AI on Jetson Nano – set up Jetson Nano, collect and annotate image data, train a neural network.
Building a Brain in 10 Minutes – how neural networks learn from data and the maths behind a neuron.
Building Video AI Applications on Jetson Nano – create DeepStream pipelines, handle multiple streams, use alternate inference engines like YOLO.
Augment Your LLM Using RAG – basics of retrieval‑augmented generation, the retrieval process, NVIDIA AI Foundations and key components of a RAG model.
Building RAG Agents with LLMs – scalable deployment, microservices, LangChain paradigms, practice with state-of-the-art models.
Accelerate Data Science Workflows with Zero Code Changes – benefits of unified CPU/GPU workflows, GPU‑accelerated data processing and machine
Introduction to AI in the Data Center – AI, machine learning, deep learning fundamentals, GPU architecture, frameworks and deploying AI workloads, planning multi‑system clusters.
I’d love to hear what you think – and what you’re reading, watching or listening to.
Drop your thoughts below.
Stay curious.
Vlad