Coinbase CEO's Top 3 Crypto Trends for 2026 + More from Davos!
These notes were generated by AI and may contain inaccuracies.
Broadcasting from the USA House at the World Economic Forum, sponsored by the New York Stock Exchange. NYSE is a modern marketplace built for scale and long-term impact.
Jason Calacanis interviews Brian Armstrong, Coinbase CEO, at Davos. Focus is on serious global regulations, particularly market structure legislation for crypto. Also involves networking and commercial meetings. Five of the top 20 global banks use Coinbase to build crypto infrastructure into their products. Meetings with country leaders about economic freedom and updating financial systems with crypto.
Partnerships include public integrations with JP Morgan and PNC Bank; others not public yet. Powering integrations with BlackRock, who wants to tokenize every fund. Financial institutions coming on-chain.
Struggled with regulators over last decade, especially under Biden administration. Contrast with Trump, who campaigned to make US the crypto capital of the world, kept promises, pushed for clear rules. 52 million Americans have used crypto, want clear rules and better financial services. Global competitiveness issue: China paying interest on CBDC, largest stablecoin issuers offshore; aim to repatriate capital to US.
Crypto bent or broke rules like Airbnb and Uber; now rules being refined. Genius Act (stablecoin bill) passed into law. Stablecoins must be audited, above board, 100% assets in short-term US treasuries (max 30-day).
Banks do fractional reserve lending, don't store all money; stablecoins with 100% reserves are safer, no bank license needed.
Silicon Valley Bank story: Run on bank after mistimed treasury allocations. In board meeting Thursday, got text to withdraw; insisted on pulling half to make payroll; bank failed that night.
Revenue from stablecoins via rewards program (not interest), per Genius Act negotiation. Rewards require activity like payments, trading, or Coinbase One subscription; pass ~100% economics to customers. Growth driver. Compares to money markets (trillions) and high-yield checking; banks nervous about tech competitor.
Some banks nervous, others leaning in. Trump wants all American businesses to win. Bank trade groups trying to undo Genius Act (passed 4 months ago); red line for industry.
Strong relationship with Circle; USDC is largest regulated stablecoin (Genius Act US, MiCA Europe). No exclusive; list others. Support Tether in certain ways (e.g., conversions), varies by country; also PayPal stablecoin. Tether not Genius Act compliant (not 100% short-term US treasuries). Likely two Tethers: US-compliant and offshore Wild West.
Like Tether team; provided dollar access in emerging markets with high inflation (70-100% yearly). Users determine risks.
Minimum listing standards: cybersecurity, no rug pull risk, legal compliance. No investment recommendations; like app stores or Amazon everything store. Disclosures and standards over ratings (ratings politicized, see The Big Short). Tried user ratings; failed due to bias (talking their book).
Tokenizing data centers, oil reserves. Top 3 trends:
- Everything exchange: all assets on-chain (crypto, equities increasingly, prediction markets).
- Partnering with Kalshi (non-exclusive); also talking PolyMarket, can launch own.
- Stablecoin payments growing.
Biggest growth: B2B cross-border payments (e.g., buying from Asia/Europe, selling in Brazil; avoids 7-day waits, high FX fees). Launched Coinbase Business for SMBs (cross-border payments, invoicing, tax, accounting); huge backlog, need to staff up. Coinbase Developer Platform like AWS for white-label wallets, trading, payments, staking, financing.
Tokenizing private shares needs company permission (vesting, retention). On-chain capital formation for private companies: register security, raise from more than accredited investors. Bill to expand accredited investor definition (SEC charged but slow). Democratizes wealth creation, lowers costs, reduces friction.
Private company demand: pent-up for shares, shady SPVs now boiler rooms charging 10% loading fees to civilians for SpaceX, etc. Unintended consequences of regulation (Sarbanes-Oxley extended private stays; Uber/Airbnb 5-year indigestion post-IPO, Instacart from $30B to $10B).
Funds on-chain: Coinbase Tokenize helps funds, real estate; democratizes access, increases demand, cuts back-office fees/settlement risk (instant on-chain). BlackRock, Apollo doing it.
Top funds in the world have publicly stated they want to tokenize every single one of their products. This is happening.
Second, third order effects of venture funds or REITs onchain: Different types of funds—some only for institutions/accredited investors, others open to retail. For retail, tens of millions of people worldwide could invest in minutes at average $100-$1,000, democratizing access. Coinbase published a report: 4 billion adults are unbrokered, lacking ability to invest in high-quality assets. This is the engine of wealth creation for capitalists. Many are stuck earning only from labor; with $100-$1,000, they might want 10% in S&P 500, Coinbase stock, Nvidia, but can't.
People use alternatives like PrizePicks (speaker uses it for Knicks parlays), but they could invest in companies they know, e.g., HR worker at LinkedIn hears about next big product, invests $1,000 for 1,000x return, life-changing.
Financial literacy component: AI agents in Coinbase app teach dollar cost averaging and tax loss harvesting.
Makes high-quality investments available, lifts people out of poverty.
In US, rise of socialism in New York, California. Speaker moved to Austin 3 years ago due to social issues, predicts California bankruptcy, trending toward insolvency, potential wealth tax or asset seizure.
Estimates: 20% of California's billionaires have left, creating $10 billion tax hole even with taxes from stayers. Biggest sell-off ever, disaster.
Voice vs. exit dilemma: Love California but abusive relationship—keeps adding burdens. Once leaving, incentive shifts to extract builders/top talent to welcoming places.
Global operators easy to relocate (speaker on planes across 4 continents yearly); family happiness key (loves Austin). Hardest issue at Coinbase: employee housing costs in California—recruiting families needing private school/house doubles cost of living, requires higher pay.
San Francisco had punitive revenue-based tax on financial services; Stripe moved out. Mark Benioff regrets supporting it (wanted to fund homeless programs).
Homelessness: Not solved by homes—they're addicted to drugs. Budget up dramatically last 10 years, services worse. More spending on homelessness increases homeless numbers. Waste/fraud massive (e.g., Nick Shirley scandals). California abuse dwarfs Minnesota; big economy, many NGOs/homeless orgs taking hundreds of millions in San Francisco alone.
Post-Coinbase 2021 IPO liquidity, continue as CEO (public company CEO cool, early journey) but use capital for big bets. Inspired by Elon (PayPal/X to atoms world). Software more forgiving than atoms (lower margins).
Met co-founders in longevity space: epigenetic reprogramming—reprogram cells to restore younger function. Hosted dinners, funded them; now board member, others invested. Company: New Limit.
Progress faster than expected: First 2-3 years, reprogrammed human cells successfully. First drug candidate into clinical trials next year.
Davos shifted from ESG/DEI/global government to brass-tacks deal-making (countries/businesses). Patina of globalization vs. nationalism.
Less ESG/DEI due to Larry Fink (new leader vibe), Donald Trump. US GDP growth, low inflation, business environment shows how to win/prosperity. Poorest do best in high economic freedom countries. Growth solves problems.
US numbers: 5.6% GDP (spectacular, not seen since COVID money-printing), unemployment 4.6% (lowest 4.3%), inflation ~2.8-2.9% (target 2%). Growth from deregulation, low-cost energy, private markets, clear rules, level playing field. Capitalism biggest win-win.
Private companies cooking, creating jobs, taxes, positive cycle.
Crypto + AI top trends; converge as AI agents need payments. Traditional finance requires human KYC; AI agents use stablecoins, crypto wallets (know your agent).
Job displacement: Techno-optimist. Early 1900s: 80% US workforce in agriculture (manual labor), now 3% after automation (over 30 years). Past views: conversations/office work not "real jobs" vs. fields. Prefer air-conditioned stressful keyboard vs. backbreaking labor.
Transition period yes, but leads to abundance. New jobs: streaming video games on YouTube, philosophical works, not packing boxes.
Examples: Robo-taxi (Wuhan/Beijing protests, limit licenses; Boston/California protect jobs). Amazon package driving/packing gone with Optimus, RoboTaxi, Waymo.
Coinbase: Same employee count (overhired then growth hiring). AI-first employees 10x knowledge workers.
Internal hosted AI model on all data (Slack, Google Docs, Salesforce, Confluence). Every team (legal, finance) uses it.
CEO uses: "What should I be aware of?" Reveals team strategy disagreements from Slacks/Docs. Reverse prompting (Tobi from Shopify): Ask AI what to think about, not dictate. Acts as mentor/coach.
Examples: Time allocation (spent 32% vs. planned 20% on X); what changed mind most last year.
Claude Co-work: Describe knowledge work goal, builds wizard-style (connected Notion/Slack/Google Docs). Analyzed: Spend more time with winning founders vs. internal team.
Andrew Feldman, CEO Cerebras Systems, building wafer-scale engine (WSE) chips for inference/training.
Demo: Wafer-sized (56x larger than B200), 4 trillion transistors. Bigger chips = faster AI results.
Cost: First one $500 million. Systems: On-premise $1-1.5 million; cloud by token (50 cents-$ several/million tokens), or monthly/yearly.
Predicted AI 6-7 years pre-ChatGPT. Surprised by speed (except Sam/Ilia). Saw new AI computational problem pressuring processors. Built 20-50x faster machine solving 75-year chip size problem (wafer-scale, many failed).
No pre-AI waypoints like Nvidia (games, crypto). Nvidia stock flat 2004-2010, graphics flat, then gamers/supercompute.
Cerebras AI-focused: Early national labs, military, pharma (pre-LLMs: protein sequencing, vision models, HPC/AI edges).
Now top compute: images, video production, model training, deep learning/deep thinking (many-threaded jobs). Deep research most compute-intensive.
Deep research uses an enormous amount of compute. When doing a deep research query, like using Claude's co-pilot application to create a podcast guest dossier with 15-20 steps—covering every podcast appearance and news item timeline—it produces a document better than anything a human assistant has made after 20 years of interviews. Human assistants previously took 48 hours for a dossier that was 20% of this, now done in under 10 minutes. Last year, assistants used it for ideas and links, cutting time from 16 to 8 hours; now, no human needed for this work. Imagine getting it in 10 seconds.
AI involves training and inference. Inference is exploding because everyone uses AI. A task kicks off threads of queries, each generating results as input for more queries, creating a cascade requiring massive compute. Traditional compute takes 10-15 or 20 seconds per query, leading to a waterfall of time. Their system delivers all answers in 4-10 seconds. Current deep research takes about 5 minutes (grab a coffee); images or 5-second videos take 90 seconds.
Speed Analogy to Internet Evolution
Like dial-up to fiber: slow internet meant Netflix delivered DVDs; fast internet made Netflix a movie studio—a fundamental change. Speed transforms AI similarly. Customer Cognition powers their coding engine with zero latency, keeping users in flow while writing code. Recently announced partnership with OpenAI (original investors, now major purchase order) for extraordinary speed improving user experience.
Google's Larry, Sergey, and team found shaving milliseconds boosts usage. Paper showed even imperceptible time cuts increase usage; just noticeable difference is 15% of current time—cut that, people use more, leave less. Paul Graham tweet: "I'd use Google half as much if ChatGPT weren't so slow." Waiting loses customers to coffee or elsewhere.
User anecdote: pays $600-700/month ($10,000/year) for Gemini, Claude, ChatGPT; runs same query across three browser windows simultaneously, burning significant compute.
Partnership with OpenAI: 750 megawatts, delivered over years via cloud infrastructure they build. Shift from chips/units to power (megawatts) started last summer; data centers now measured by power, not square footage—power is the limiting constraint.
Optimal Data Center Locations
Cheapest power: hydro, then natural gas. Natural gas abundant in West Texas, Wyoming, Caribbean, Ghana; geothermal in Nordics. Uses byproduct flare-off gas (previously burned; Bitcoin miners utilized it).
Hydro: known for data centers; water-cooled systems efficient. Early adopters of production AI water cooling (TPUs too). Cooler water better (Alaska, Canada ideal); use chillers or groundwater. Closed-loop systems recycle water—heat pulled off chips, water chilled, pumped back. Not comparable to golf courses' water use; no chemicals, water undamaged. Misperceptions about AI water usage.
Frustrations: hyperscalers mishandled rural communities (cheap land, abundant power) by poor communication, cutting power company deals raising local rates over 20-30 years. Should have been good citizens: big taxpayers, build schools (rounding error cost), create construction jobs.
Avoided pitting communities against each other for tax discounts. Microsoft example: guarantees data centers won't increase utility costs—thoughtful for national champion.
Ideas: home solar, batteries (e.g., Michael Dell's son project) store cheap excess power for peak demand (duck curve), easing grid. Could enable free energy with data centers.
US grid decrepit (40-50 years underinvestment), behind China. Home storage reasonable: reduces grid load, adds resilience (e.g., California outages 2 days for wind/hot/cold).
France subsidies: free lifetime energy for living near nuclear plants.
Nuclear Power
Small modular nuclear: right thing, efficient over 20-30 years (beats all others), but upfront expensive. Not for data centers next 3-4 years. Would build if possible. UAE building modular nuclear data centers.
Elon Musk: data centers/chips in space—easy cooling, effective solar. They've researched; betting against Elon long-term unwise, but timing hard (he's never wrong, frequently late). Space solar smart: closer to sun, thinner atmosphere. Challenges: vacuum cooling, satellite communication, data back to Earth (past satellite internet glitchy/delays; low-orbit better, but work needed). 8-10 year horizon.
All options worth pursuing; not overbuilding.
Early in AI compute demand: tiny enterprise adoption changing workflows. Consumers use 6-8 times/day now; imagine 100x, all devices autonomous, every engineer as coding co-pilot. Demand surges: better models, more users, frequent use, more compute per usage—inference dominant.
Computer Architecture Balance
Three pillars: calculation speed, memory storage, I/O delivery. Balance essential—fast calc bottlenecked by slow storage/IO. Senior architects/CTO design holistically (like car: fast but can't turn useless except drag racing).
GPUs: much memory capacity but slow—huge inference bottleneck (xAI spent $20B on Grok for fast inference answer). Constant tradeoffs: power use, memory speed/capacity/cost.
Memory shortage: demand signals exploded (6 to 18 months ahead); supply chain confused, same production as 4 months ago. Takes 18 months to digest; prices high (cycles every 6-8 years). GPUs chew HBM (DRAM flavor), less for consumers.
Geopolitics real. US ahead in chipmaking (Santa Clara: Intel, AMD, Nvidia, Grok team, ARM, etc.—6/10 top teams). Expertise from building high-speed chips iteratively.
China behind there, but ahead in open-source models (top-down: modernized grid, huge power). Industrial adversaries; winner-take-all: recursive systems amplify leads (100x developers, knowledge workers, biotech). Iteration speed magnifies small edges—race critical.
Davos full of politicians; Trump focused (vs. Biden team demonizing Silicon Valley). Objectively, how do you think
How do you think objectively the Trump administration is doing on their AI policy and the support they're giving the AI industry? Even on a lot of fronts they're doing really well. Previous administration made a mistake keeping chips from allies. UAE is clearly an ally, modern Arab nation, source of peace, made peace with Israel early on, huge Western influence. Kept chips from them. Like KSA, Kingdom of Saudi Arabia to move in the same direction, kept chips from them. Made a hierarchy that made the Danes feel second rate, said "You are a number two friend." Bad idea. Should be empowering allies. Previous administration didn't understand that. Trump did a good job. Want those nations and their institutions using our technology and investing in the US. Previous administration had a CFIUS organization in Treasury that was difficult to work with. All that's much improved. They were unclear, not communicative, impossible to deal with. Super important to be the standard, so all that energy goes back into our ecosystem, development on top of us, recursive system. Look at Huawei with 5G, networking up against Cisco and national champions, they ran the table in a lot of countries, clobbered us in Africa and developing world. Now those places have spyware. Real issue. Those were areas where this administration did the right thing.
Energy Policy
Energy. One thing that kills a company trying to grow quickly is dealing with different regulations in each of 14 localities for data centers. Brutal. Don't want 17 lawyers figuring out local rules. Effort to get reasonable laws across the board is smart. Getting money to improve the grid across the nation would be helpful. Work with Department of Energy under Chris, Genesis program, equivalent of Manhattan project for AI. Need to think among researchers not how to get 10 or 20% faster, but use AI to increase rate of research by 5x, 10x, get things that impede government out of the way.
China Challenge
China's a sticky problem. Not sure I agree with current push to allow selling H100s there, but reasonable to disagree. Not clear-cut, hard problem, lots of different views.
Alliances and Reliability
Canada just made a strong alliance with China. Trump administration can improve by maintaining alliance with neighbors so they trust us. Japanese feel we're not most reliable partner. Canada feels not reliable because of tariff issues, military issues, consistency important in policy. Canada has huge raw material exports: wheat, lumber, stuff we import or they take elsewhere. Have to be aware of realpolitik, they have to sell raw materials, China big buyer. Nations are proud, constant attacking gives leaders ability to say China is courting us, going to invest, build ports with them. Chinese ownership of major ports and shipping crazy, own world's large shipping routes, Belt and Road strategy. In third world, see BYD cars everywhere. In Mexico City, every guard of BYD. Government subsidizing BYD by 30, 40, 50%. Goal to put Germans, English car manufacturers out. BYDs getting footholds in Europe. Who's going to buy $40-60k Beamer, Volvo, Audi when $20-40k BYD? Price dumping, tariffs meant to protect against that. Subsidizing at finished product benefits whole supply chain: battery maker, transmission maker.
Microsoft, Uber, Coinbase, Meta, Google have more employees four or five years ago or flat. Young people unemployment hitting 10, 20% among college age demographics. Is it AI, entitled kids without work ethic, overfunding and tech indigestion? Startups most resourceful, AI first, building agents. No world without AI displacement. Not why displaced now, but 100% coming. Not an issue today because people let go are middle management, delayed impact of good SaaS tools. Ability to extend reach as leader, scope much bigger. Middle management role to move information, manage small teams, keep on track, shrunk in value. Halfway there with AI, but not yet. Ballooning of jobs, Zuckerberg and Satya look and say competition intense, what are these waves of people doing, slowing us down. Flattening organizations. Down the road, whole categories vastly more efficient, need less people. Rest in peace Scott Adams, creator of Dilbert, giant ridiculing corporate America, that exact layer gone. Great he got to see it.
Nuclear Energy Critique
If you see Germans, ask why they turned off their nukes. Turned off three of six nuclear reactors, decided to import natural gas from Russia. Bad, depended on Russia. All because of Davos, WEF, caught up in virtue signaling, never looked first principles at how safe nuclear is compared to burning fossil fuels. Nuclear safe, can make safer. In Japan last week, putting in new nuclear reactors, got over Fukushima, realized mistakes like putting below sea level, not making again. Nuclear way to go, get better at it.
Welcome back, grifting interviewing top CEOs at Davos, World Economic Forum, first time here, great time, hundreds interview requests, doing half dozen. Jake Lucerarian, friend of pod, CEO co-founder Gecko Robotics, almost decade. Build purpose-built robots to inspect ships, bridges. Started long before ChatGPT AI revolution. Have you put AI into them? Previously straightforward, inspect bridge, now can think on own, more adaptable? Changed title to chief grifting officer in Davos. Caught tail end of DEI, had to learn different language, everything super precious, now with Trump here, brass tacks, doing business, negotiating, less performative. CEOs talking AI, but conversations about ROI from AI. Gap in AI: information and data sets for ROI, productivity gains for large infrastructure asset owners like energy, mining, manufacturing. Turning into business conference, astounded by inbound pure business capitalism.
Military and Defense Applications
30% business defense, Department of War. Use technology to fight speeds of development of China in manufacturing. Understanding quality and welds bottleneck for US, manufacturing forges 100 years old. Robots help manufacture components of submarine, expedite destroyer turnaround to patrol borders, deter conflict. Speed up decision process, accurate. Admiral Houston: 90% improvements to speed up manufacturing. Companies like Anduril working with us, Palmer Luckey.
Energy Sector Growth
Biggest growth: large energy companies, power companies. Hyperscalers focusing on capex for infrastructure. Play game with GDP driver companies, AI native or robot native first to support AI initiatives, build data infrastructure to layer AI models. Started company 13 years ago with premise data matters. Robots have sensors, arrays inspect metal, fabrication, seams of submarine, measure perfectly. Diagnose health of built world: bridges, dams, submarines. Centralize information, layer operational data from sensors, make decisions to extend useful life of asset, push harder to produce more, create cleaner more barrels per day at lower costs. Robots dedicated to business problems: making barrel, kilowatt, ship out of dry dock faster.
Future Roadmap and ROI
Not building humanoids yet. Biggest purchasers of Optimus robot. Folding laundry, cleaning dishes not high ROI. $20/hour use cases. US best at using robots for unfair advantages in oil/gas, power. Between Tesla, Figure, Boston Dynamics, need application layer, operational excellence. Nervous system pulls info from robots, build AI models, take actions back to real world. Roadmap: assess state of built world health, decide repair actions, automated welding solutions, feedback to create foundation model for best welding. Build robots to identify and solve highest ROI problems: manufacturing new assets, operate/maintain existing. Reduce hazardous work hours, extend asset life, increase capacity/production, catch failures. Easy to underwrite problems.
Human-Robot Collaboration
Humans in loop for time to come. One welder supervising 10 robots. Experience, subject expertise ensure robot understands ramifications. Teleoperations with mobile robots, humans in AC building operating, building data to train foundation models. Reduce risk in deep sea welding, bridge climbing. Reduce barrier to entry for jobs taking 10,000 hours: make attainable in months. Shortage of welders, inspectors, trades. Augment: take Home Depot employee, in months make $100-150k running robot safely. Key unlock: get robotics in field, fail fast, rapid prototype. Manufacturing big issue. Over next 5 years, best in world at ROI from robots.
Pittsburgh Editorial
Wrote editorial: live in Pittsburgh, 1930 more millionaires than New York, made 70% world's steel. Economy changed, doesn't anymore. Insights and
Companies like those at Davos remain the backbone of our economy. Inspired by the industrial revolution, where steel was invented, manufactured, and distributed to create infrastructure fueling energy and manufacturing sectors, robotics plays a similar role today. Robotics collects information and data sets to build infrastructure for AI models to deliver expected returns.
Robotics forms the foundation for massive returns in key sectors. Automation can eliminate low-value middle managers, but serious ROI requires real-world deployment like robo-taxis or self-driving tech. The Davos forum and Silicon Valley represent bubbles focused on internet worlds, ignoring sectors like energy, metal manufacturing, mining, and defense, which were once taboo.
Silicon Valley has exhausted SaaS software opportunities, with pitches shifting from novel ideas to the 50th or 15th company in a vertical, running out of targets.
Humanoid robots' first demos, like folding laundry, miss critical applications. Living in Pittsburgh reveals a different world from Silicon Valley. Talking to energy company CEOs shifts perspectives. Advice to startups: focus where the real problems are, not conversations—exit the ecosystem to solve important issues.
Boston Dynamics did backflips a decade ago but lacked LLM, vision model, or world model. Now, with these, robots can be instructed, e.g., "lay some bricks," then access web resources like YouTube videos, manuals, and device histories to perform tasks without specific training.
"I know kung fu. Boom. It just does it."
Such capabilities could arrive in a three-year timeframe. Big investments include SoftBank and Nvidia's $1 billion into Skilled AI (now $14 billion valuation), founded by Deepo in Pittsburgh, developing robot brains.
In energy, defense, and similar sectors, no existing video corpora or data sets exist. Solution: equip humans with GoPro cameras and sensors for training data by modeling human actions.
Few customers pay for low-ROI training like arm sensors, as energy or manufacturing firms won't spend $10 million/year. Instead, collect data by solving critical problems on infrastructure and assets, e.g., walking Manhattan-sized refineries.
Refinery inspections today involve humans dangling 100ft up on ropes, taking pictures and using sensors. Robots with fused sensors create novel datasets not on the internet, giving Gecko a significant advantage in world-building.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
