Sam Altman: Intelligence Got 100x Cheaper in 2 Years. His Next Bet Is $5 Trillion

OpenAI CEO Sam Altman said the price for a given level of intelligence falls by about 10x per year, and that the company's Deep Research feature handles about 5% of all tasks in the economy today. Altman, speaking on The Times Tech Podcast in London, also outlined plans to scale a $500 billion infrastructure project to $5 trillion, and noted a warmer relationship with Washington compared to two years ago.

Intelligence has a price, and the price falls on a schedule. “The price for a given level of intelligence, once it’s achieved, falls by about 10x per year.” Sam Altman said that to The Times, and the single number funds everything else in the conversation: a $500 billion infrastructure project he wants to turn into $5 trillion , a Chinese rival that barely registered inside OpenAI, and a noticeably warmer Washington than 2 years ago. I listened to the full Times Tech Podcast, recorded in London with Katie Prescott and Danny Fortson, so you can skip it. Here are the 10 takeaways that matter. t ogether with Outskill: Intelligence is getting 10x cheaper every year. The people who win are the ones who know which tools to point it at, and hand over real work. This Sunday, a free live 3-hour workshop covers the 15 AI tools that matter right now : ▫️ Pick the right tool https://links.outskill.com/AICOAUG5 for research, writing, design, data, code, or automation ▫️ Build your own AI co-worker https://links.outskill.com/AICOAUG5 that runs 24/7, even while you sleep ▫️ Ship apps https://links.outskill.com/AICOAUG5 , dashboards, and workflows with zero code Usually $395, completely free for readers. Live Sunday, 10 AM EST. 1. What Took Weeks Now Takes 20 Minutes His Own Users Proved It First Altman didn’t need a slide deck to make his case. His users made it for him, over one weekend. “People like a lot of people who are, maybe they were even recent AI skeptics were saying things like, I can now do things that would have taken me like many days or even weeks of work. AI can do like it in 20 minutes and it can do a bunch of them in parallel." OpenAI shipped Deep Research https://theaicorner1.substack.com/p/autoresearch-playbook-agent-optimization-loops-2026 7 days before this interview aired. Altman spent the weekend reading reactions instead of writing a press release, and the reactions that stuck came from people who’d doubted the technology to begin with. His own framing is a projection, not a snapshot. Run that same rate of change forward another 2 years, then a full decade, and he expects the bottleneck stops being what the model can do. It becomes what a person can think to ask for. If your team still measures a research task https://www.the-ai-corner.com/p/ai-inference-engineering-playbook-2026 in days, you’re pricing in a version of AI that expired 7 days before this podcast aired. 2. One Feature Already Handles 5% of the Tasks in the Economy Altman puts a number on Deep Research that most CEOs would keep off the record. "My vibes based estimate is that does about 5% of all tasks in the economy today, one feature, 5% of all the tasks in the economy today." He’s upfront that it’s a vibes-based estimate, not a study. What makes the number land is the task list behind it, a named set straight from the interview: ▫️ Summarizing an entire field of research https://theaicorner1.substack.com/p/karpathy-autoresearch-agent-improvements ▫️ Comparing baby cribs against a buyer’s own constraints ▫️ Drafting a report at consultant quality https://www.thevccorner.com/p/mckinsey-pyramid-principle-claude-investor-documents-2026 ▫️ Running a full financial analysis https://theaicorner1.substack.com/p/claude-for-investing-4-level-system None of those are demo toys. They fill a junior analyst’s week. Altman is specific about sequencing too: knowledge work https://theaicorner1.substack.com/p/the-salary-that-disappeared-6-roles absorbs the impact first, well ahead of physical work, since robotics is still catching up behind it. If your business runs on research, summarization, or first drafts, that 5% figure already happened, inside the 7 days before this interview. 3. The Price of the Same Intelligence Falls 10x a Year This is the number that funds every other decision in the interview. "Our observation in our field is that the price for a given level of intelligence, uh, once it's achieved falls by about 10 X per year. So a hundred, a hundred, a hundred X in the last two years." Altman reaches for his own comparison to make the curve legible. Moore’s Law doubled transistor counts every 18 months and reshaped the global economy on that slope alone. This curve runs steeper, and it’s the same curve behind Anthropic cutting the price of frontier intelligence in half https://theaicorner1.substack.com/p/claude-opus-5-playbook-benchmarks-pricing-effort-dial-2026 . That’s also why DeepSeek’s cheap, capable model https://theaicorner1.substack.com/p/kimi-k3-playbook-benchmarks-cost-setup-2026 didn’t rattle his research team the way it rattled markets. Altman calls it, in his own words, not a big research update internally, even while crediting the DeepSeek team for a couple of smart product choices: showing the model’s chain of thought, and a generous free tier. Don’t price a 2026 product against today’s inference cost. The unit economics https://theaicorner1.substack.com/p/token-doomed-unit-of-sale-ai-pricing of anything AI-powered improve on a clock, and that clock runs faster than most planning documents assume. If your roadmap still prices this year’s inference cost, these break down where the curve goes next: ▫️ Anthropic just cut the price of frontier intelligence in half https://theaicorner1.substack.com/p/claude-opus-5-playbook-benchmarks-pricing-effort-dial-2026 ▫️ You are overpaying for intelligence. Grok 4.5 just proved it https://theaicorner1.substack.com/p/you-are-overpaying-for-intelligence ▫️ The Real Reason AI Costs Keep Rising https://theaicorner1.substack.com/p/token-doomed-unit-of-sale-ai-pricing 4. Why Altman Says He’s Still Afraid of Becoming AI’s Napster Asked whether OpenAI could get overtaken the way Napster got replaced by cheaper, faster followers https://theaicorner1.substack.com/p/where-ai-moats-live-now , Altman answers head-on. "Of course I'm worried about it. You got to wake up every, the only way to not have that happen is to wake up every day, worried about it." He follows the fear with a check that keeps OpenAI spending even as rivals ship cheaper models. He says he’d welcome it if a competitor convinced the entire industry that massive compute clusters are unnecessary https://www.the-ai-corner.com/p/marc-andreessen-ai-moat-not-the-model-2026 . He keeps funding the infrastructure anyway, treating that hope as a bonus, over a plan. The lesson travels past AI labs. Comfort is the bigger competitive risk https://theaicorner1.substack.com/p/you-are-overpaying-for-intelligence , ahead of whichever competitor everyone can already see coming. 5. Stargate Is $500 Billion Today. He Wants $5 Trillion Next. The scale Altman describes stops sounding like a technology budget and starts sounding like a country’s infrastructure plan. “Stargate is a $500 billion project to build a very large training inference system. It sounds crazy big now. I bet it won’t sound that big in a few years. If we get to do this again, you’ll be raising $5 trillion for a cluster.” OpenAI is currently raising $40 billion https://theaicorner1.substack.com/p/groq-nvidia-deal-origin-story specifically for Stargate, led by SoftBank, a figure Altman confirms is a component of the larger $500 billion build https://theaicorner1.substack.com/p/cerebras-series-a-deck-teardown-ipo-2026 , not the whole company raise. He also confirms conversations toward a Stargate Europe had already started the week before this interview. He doesn’t treat that scale as guaranteed to land smoothly. He expects booms and busts along the way, and says he’d happily buy someone else’s overbuilt infrastructure at 10 cents on the dollar https://theaicorner1.substack.com/p/ai-semiconductor-pullback-playbook-micron-nvidia-2026 when that correction comes. The compute race moves fast, and it rewards whoever tracks it closely. Start here: ▫️ Cerebras bet against the GPU. It just IPO’d at $56B https://theaicorner1.substack.com/p/cerebras-series-a-deck-teardown-ipo-2026 ▫️ The AI trade just lost $1 trillion. Here is where the value went https://theaicorner1.substack.com/p/ai-semiconductor-pullback-playbook-micron-nvidia-2026 ▫️ Jonathan Ross Turned a 3-Week Phone Call Into a $20 Billion NVIDIA Deal. Here Is How He Did It. https://theaicorner1.substack.com/p/groq-nvidia-deal-origin-story 6. Governments Are Already Asking for Their Own Stargate Altman expected small requests from governments on this trip. He got bigger ones back. "I was pleasantly surprised on this trip... There's some governments that are ready to like buy big pieces of AI infrastructure." The interviewer puts a contrasting number on the table earlier in the same conversation: the UK’s own commitment sits at roughly 800 million pounds for a national supercomputer, a fraction of what Altman is describing governments now asking him for directly. His fallback for governments that won’t write a check that size is direct: buy AI as a cloud service instead of owning the infrastructure, on the bet that the world stays open enough to sell it to them. Sovereign AI infrastructure https://www.the-ai-corner.com/p/the-us-government-switched-off-anthropics just became a live budget line, not a talking point. Watch which governments move from conversation to signed commitment https://www.the-ai-corner.com/p/dario-amodei-zeroth-world-davos-2026 first. 7. Why Altman Says Trump Is Easier to Build Under Than Biden Was Altman states the comparison plainly when Danny Fortson puts it to him directly. “President Trump has such a different opinion on building things, and permits, and power, and manufacturing in the US.” In a separate answer, he calls the Biden administration not easy on the infrastructure front specifically, naming permitting as the sticking point. He’s careful to add he didn’t agree with every position from either administration, and treats that as ordinary rather than a contradiction worth defending. The tech and the infrastructure are inseparable in his framing. You can’t train frontier models without the power and the permits to build the data centers first, which makes permitting speed a strategic variable for any founder whose roadmap depends on physical infrastructure https://www.thevccorner.com/p/how-trumps-2025-tariffs-impact-startups . 8. Inside Silicon Valley’s Vibe Shift Altman names the mood in the Valley directly, and walks back his own first word choice mid-sentence. “The word that came to mind for the last administration was hostile. I think that’s a little bit too strong, but they were not friendly to tech or business. It’s a very welcome breath of fresh air.” He describes a wave of tech companies giving to the inauguration fund and to the presidential library, and frames it as founders finding whatever they can build alongside a president on, over agreeing with him on everything. The upside he names is concrete: a shot at rebuilding semiconductor fabrication https://theaicorner1.substack.com/p/jensen-huang-nvidia-founding-algorithm , robotic factories for data centers, and new energy generation inside the US, categories the country had mostly ceded. Political alignment in tech right now reads as transactional https://www.thevccorner.com/p/sam-altmans-10-rules-for-the-ai-era . The founders donating are buying permitting speed, not endorsing a platform. 9. Altman on Elon Musk Today: “No, I Don’t” Two co-founders, one active lawsuit, and a question that gets a shorter answer than expected. “Do you miss him as a friend? No, I don’t. I think he’s really changed.” Musk is currently suing OpenAI https://theaicorner1.substack.com/p/musk-altman-openai-trial-dossier-court-filings-2026 over its shift toward a for-profit structure, a case Altman jokes he’s lost track of the exact version of. Asked when he last spoke to Musk, in person or by text, his answer sits close to nothing. He still credits Musk with accomplishing extraordinary things and calls him a genuine early inspiration. The friendship is what changed, in Altman’s own account, even as the respect for what Musk built stays intact. The Musk and Altman story didn’t start with this podcast. Catch up here: ▫️ Elon Musk and the Outer Limit of Vertical Integration https://www.thevccorner.com/p/elon-musk-xai-spacex-vertical-integration ▫️ Elon Musk’s Most Important Interview in Years https://www.thevccorner.com/p/elon-musk-wtf-interview-10-big-ideas-for-founders-investors ▫️ Elon Musk’s 2007 Interview Predicted the Next 18 Years https://www.thevccorner.com/p/elon-musk-2007-predictions 10. The Paul Graham Line Altman Still Can’t Shake The interview closes on a decade-old quote he still hasn’t found a clean answer for. “It is not what I wake up in the morning thinking about. I really don’t. But I can’t argue that I ended up in a fairly influential position. I don’t know how to square those.” The line in question comes from a New Yorker profile https://www.thevccorner.com/p/replit-pitch-deck-seed-2016-story-playbook-2026 written roughly a decade earlier, quoting Paul Graham calling Altman extremely good at becoming powerful. He accepts the outcome. He calls the intent behind it a mystery to him. He adds a detail that undercuts the image of a man scheming his way up: he says he never pinches himself before meeting a head of state, since people adapt to anything, good or bad, faster than they expect. Power, in Altman’s own telling, is a side effect he stopped trying to explain. Believe that framing or don’t. Either way, it’s the frame he chose to hand you on his way out the door. The AGI Economics Playbook The thesis in one line: intelligence gets cheaper on a fixed schedule, and the winners build infrastructure, government relationships, and products around that curve instead of today’s price. ▫️ Founders: stop budgeting AI costs at today’s price. A 10x-a-year curve means your unit economics improve on a schedule, so build the product you can’t afford yet, over the one you can afford today. ▫️ Investors: watch permitting speed and government infrastructure deals as closely as model benchmarks. Altman is choosing jurisdictions and administrations that move fastest on power and permits, and that belongs on your diligence checklist now. ▫️ Operators: re-audit which of your team’s tasks already sit inside that 5% number. That was one feature’s output after 7 days on the market, not a ceiling. ▫️ Everyone else: sovereign AI infrastructure is becoming a national budget line, over a research grant. Track whether your own government is negotiating for compute access or getting left out of that conversation entirely. The 5 Principles to Steal Price for the curve, not the sticker price. Intelligence gets 10x cheaper every year. Plan the product you can’t afford yet. Stay afraid on purpose. Altman’s answer to the Napster question is to worry every single day, over relaxing after a good quarter. Infrastructure and permits are strategy now. The company and the country that build power and data centers fastest set the pace for everyone else. Alliances can be transactional without being dishonest. Altman doesn’t agree with every position from any administration. He works with whichever one builds faster. Let the outcome speak instead of the explanation. Altman never fully answers whether he chased power. He just keeps building the thing that keeps handing it to him. Intelligence got 10x cheaper again this year. Most teams are still pricing last year’s model. The ones who aren’t are building Stargate-sized bets right now. FULL PODCAST: If this breakdown saved you an hour, send it to one founder or investor who needs it. They will thank you later. Keep reading Build for the infrastructure race ▫️ Jensen Huang handed you the AI roadmap https://www.the-ai-corner.com/p/jensen-huang-ai-roadmap-10-moves-2026 ▫️ OpenAI’s $122B masterclass: 10 takeaways from Sarah Friar https://theaicorner1.substack.com/p/openai-sarah-friar-122b-masterclass-10-takeaways-2026 More from the people building this ▫️ Sam Altman: 3 months of work now takes 7 minutes https://theaicorner1.substack.com/p/sam-altman-startup-school-2026 ▫️ Jensen Huang Bet Wrong, Told the Truth, and Built a $5 Trillion Company https://theaicorner1.substack.com/p/jensen-huang-nvidia-founding-algorithm Power, politics, and the people who chase it ▫️ Replit Was Rejected by YC 3 Times. Then Paul Graham Called Sam Altman Personally https://www.thevccorner.com/p/replit-pitch-deck-seed-2016-story-playbook-2026 ▫️ 89 Best Startup Essays by Top VCs and Founders Paul Graham, Naval, Altman, Khosla & More https://www.thevccorner.com/p/best-startup-essays-vc-founders