AI analysis from the edge of adoption.
Azeem Azhar's Exponential View examines exponential technologies through the live evidence of AI adoption: agents, inference, local hardware, media disruption, and changing intellectual habits. It is analytical, fast-moving, and unusually practical about what these systems already do.
PRESS PLAY
Azeem Azhar's Exponential View is an analytical show about AI and exponential technology as lived reality. It is interested in the machinery, but also in the habits and institutions that machinery changes. Recent episodes track the move from training-era AI to inference-era AI, where models are called constantly by agents, workflows, devices, and companies. That shift gives the show its urgency. A discussion of NVIDIA becomes a way to read the future of token budgets, agentic systems, and enterprise demand. An episode on Apple becomes a reconsideration of whether the company is behind in AI, or quietly advantaged because its hardware is where advanced users already work. A segment on AutoResearch turns open-source software into a broader argument about cheaper hypothesis testing and automated scientific loops. The show is strongest when it treats these tools as things already altering behavior. Azeem Azhar describes AI chief-of-staff systems, parallel channels, research agents, argument engines, quiet thinking practices, and the uneasy boundary between delegation and deskilling. That gives the analysis a grounded texture. It is not only about what OpenAI, Anthropic, NVIDIA, or Apple might do next; it is also about what happens when one early adopter starts using tens of millions of tokens a day and sees the economics make sense. The guest conversations widen the lens. Rohit Krishnan helps explore a future of vast agent populations coordinating online. Nicholas Thompson frames journalism through creators, search traffic, trust, and AI's role as both competitor and collaborator. Other exchanges probe whether people are using AI as a tool or surrendering too much judgment to it. The tone is informed, restless, and sometimes skeptical of its own conclusions. Listeners get frameworks, caveats, and concrete examples rather than simple predictions. The show suits people who want to understand AI through markets, work, media, cognition, and power. Its central beat is clear: exponential technology stops being abstract once it changes what a single person or institution can actually do.
Made for: For founders, operators, technologists, analysts, and policy-minded listeners who want practical insight into how AI is changing work, markets, media, and institutions. It also suits heavy AI users trying to think clearly about leverage, judgment, and deskilling.
What sets it apart: The show combines macro technology analysis with specific accounts of personal AI workflows, from AutoResearch loops to AI chief-of-staff systems. It connects chip economics, software behavior, and human cognition without treating any one layer as the whole story.
How will the future unfold? What is the impact of AI and other exponential technologies on business & society? Join Azeem Azhar, founder of Exponential View, on his quest to demistify the era of exponential change.
Azeem turns a near-100-million-token day into a stress test for your judgment.
Press play if you want
Short reviews from the PlayNext desk, based on the episodes we processed.
A grounded look at NVIDIA's move from training infrastructure toward an AI inference economy.
NVIDIA's GTC gives the episode a practical way into AI inference: the point where models stop feeling abstract and start producing recommendations, summaries, images, and other useful work.
A clear rethink of Apple's AI position through the hardware that makes everyday model use practical.
Apple's AI position looks different when the focus moves from leaderboard optics to the machines people actually use. The episode frames the company as easy to underrate if success is measured only by token sales, data centers, or frontier research, while Macs remain part of the…
A thoughtful look at how AI changes the work of thinking when it becomes part of the reasoning process.
The episode frames AI as a pressure test for cognitive control, focused on what happens when an argument engine built from about 100,000 words of writing starts shaping the writer's own reasoning.
A grounded look at one person's AI system, and the judgment questions that appear when delegation runs in parallel.
A personal AI operating system comes into focus as a knowledge dashboard drawing on internal systems, a CRM, research tools, X, email, and other sources.
A sharp conversation about using AI without surrendering the abilities that make judgment possible.
AI's promise of assistance meets a harder question here: what happens when convenience starts to weaken the abilities people rely on for judgment?
A grounded conversation on agentic AI, personal-scale usage, and the guardrails a trillion-agent economy would need.
Azim and Rohit frame the trillion-agent economy less as a grand prediction than as an infrastructure challenge. If agentic systems are browsing, transacting, and completing tasks for people, the pressure moves to coordination, security, and verifiability.
The future, tested against evidence.
Future systems, argued in long form.
Innovation, measured by who it actually reaches.
Greg Williams joins Exponential View, discussing the value of clear thinking and frameworks in a commodified media landscape.
Andrej Karpathy’s 600-line AutoResearch tool has 57,000 GitHub stars and automates machine-learning experiments.
["Azeem Azhar's Exponential View ties Jensen Huang's NVIDIA GTC comments to AI inference, OpenClaw, and the token budgets that now shape daily AI use.", "He says NVIDIA is a $4 trillion company with about a trillion dollars of committed orders for Blackwell and Vera Rubin through 2027, even as Google TPUs, Amazon Trainium, and AMD compete.", "He explains pre-fill versus decode, arguing that decode is sequential and bottlenecked by memory bandwidth rather than raw compute.", "Azeem uses his own Mini Arnold system, which hit 870 million tokens in a day, to argue that firms need explicit token budgets and that Jensen Huang's half-of-salary rule for engineers is increasingly relevant."]
Azeem Azhar says Apple looks weak by AI benchmarks because it is not selling tokens, building data centers, or leading research, even though he and others run ChatGPT and Claude on Apple devices every day.
Azeem Azhar's Exponential View frames a solo essay around whether AI improves his thinking or pushes him toward "cognitive surrender." He distinguishes "cognitive offloading" from surrender, cites an SSRN paper by Shaw and Knave, and uses remembered phone numbers as the harmless example.
Azeem Azhar of Exponential View describes running "R Mini Arnold" (RMA), an AI orchestrator built on OpenClaw that manages six to eight sub-agents from a Mac Mini with 64GB of RAM in his studio.
In the debut episode of his new AI Vistas series, Azeem Azhar of Exponential View convenes Duke bioethicist Nita Farahani, Scripps Research Translational Institute founder Eric Topol, and engineer-investor Rohit Krishnan, moderated by The Atlantic's Nick Thompson, to debate whether humans remain in control of their AI tools.
Azeem Azhar hosts Rohit Krishnan, writer of the Substack Strange Loop Canon, to discuss what he calls the 'trillion-agent economy.' Azhar reveals he nearly hit 100 million tokens of personal AI usage in a single day, while Krishnan describes his own token usage jumping from 17 billion in Q4 2025 to 50 billion in January.
Atlantic CEO Nick Thompson tells Azeem Azhar that search is the most pressing of four threats to journalism, with about 40% of Atlantic traffic coming from search and a roughly 20% decline in it.
Azeem Azhar discusses Apple's AI announcements, Google's declining ad traffic, and Sam Altman's concept of a gentle singularity.
Box CEO Aaron Levie discusses the transformative impact of AI on workplace technology and management over the next few years.
Azeem Azhar discusses with AI expert Kai-Fu Lee how generative AI is perceived differently in Asia compared to the US and Europe.
Nathan Benaich discusses AI trends and predictions for 2025, including the impact of recent Google announcements and competition in AI semiconductors.
Azeem interviews Dylan Patel about AI surprises in 2025, focusing on infrastructure, investment, and bottlenecks.