Coinbase CEO's Top 3 Crypto Trends for 2026 + More from Davos!
Speakers
Brian Armstrong
— CEO of Coinbase
Andrew Feldman
— CEO, Cerebras
Jason Calacanis
— Angel Investor / Founder, LAUNCH
Jake Loosararian
— Founder and CEO, Gecko Robotics
Summary
Jason Calacanis hosts three interviews from the USA House at Davos. Coinbase CEO Brian Armstrong lays out his three biggest crypto trends for 2026 - all assets moving on-chain, prediction markets and stablecoin payments - and discusses the Genius Act, bank partnerships, tokenization and AI agents paying with stablecoins. Cerebras CEO Andrew Feldman argues AI compute demand is still in its infancy, explains why memory prices will stay high for about 18 months, and discusses power as the binding constraint on data centers, nuclear, China and BYD. Gecko Robotics CEO Jake Loosararian describes robot-collected physical-world data as the foundation for AI return on investment in energy, manufacturing and defense.
- Armstrong: Genius Act stablecoins must hold 100% short-term Treasuries; B2B cross-border stablecoin payments are the fastest-growing area.
- Five of the top 20 global banks build crypto infrastructure on Coinbase; BlackRock and Apollo plan to tokenize all their funds.
- Armstrong's top crypto trends: everything on-chain, prediction markets, stablecoin payments; AI agents will transact with stablecoins.
- Feldman: AI compute demand is early - enterprise adoption tiny, inference usage and compute per query rising, no overbuild.
- Feldman: panic ordering (6 to 18 months of demand) keeps DRAM and HBM prices high for roughly 18 months.
- Power is the limiting constraint on data centers; hydro and natural gas are cheapest, nuclear is right long-term but 3-4 years out.
- Feldman sees subsidized BYD EVs undercutting German automakers; middle-management layoffs are SaaS-driven, AI displacement comes later.
- Loosararian: Gecko's robots build physical-world datasets for energy, manufacturing and defense; 30% of its business is defense.
Ideas
Institutional crypto adoption is now irreversible.
Crypto adoption is massive and effectively irreversible: roughly 500 million people globally and about 52 million Americans have used it, Bitcoin was the best-performing asset class of the last decade, and the largest financial institutions are now integrating it - five of the top 20 global banks (including JPMorgan and PNC) build their crypto infrastructure on Coinbase, BlackRock wants to tokenize every one of its funds, and a top-10 global bank CEO told him crypto is his number-one, existential priority. With the Genius Act now law and the Trump administration pushing clear market-structure rules to make the US the crypto capital of the world, he argues it is foolish to pretend this is not happening; crypto and AI are the two most important technology trends and will converge because AI agents need on-chain payments.
Regulated stablecoins driving payments growth.
Stablecoins are the core growth engine of crypto. Under the Genius Act - now law and, in his view, a red line that will not be undone despite bank trade groups trying - US-regulated stablecoins must hold 100% of reserves in short-term (about 30-day) US Treasuries, which makes them about the safest instrument available and lets a full-reserve product be offered without a bank license; Coinbase passes roughly 100% of those economics to holders through a rewards program, which is a big driver of growth. Stablecoin payments are growing like crazy: the biggest growth area over the last year has been B2B cross-border payments, replacing 7-day settlement and high FX fees, and AI agents will increasingly transact with stablecoins and crypto wallets because the traditional KYC-based financial system is built around humans. He sees this as another flavor of the money-market-versus-bank-deposit shift rather than something that destroys lending markets.
Power availability constrains AI data-center build-out.
Power, not square footage or chip count, is the binding constraint on AI data centers: large deals are now sized in megawatts (the Cerebras-OpenAI cloud deal is 750 MW delivered over several years) because how much power can be delivered is what limits large deployments. The cheapest power in the world is hydro, without question, followed by natural gas, so build-outs go where gas is abundant - West Texas, Wyoming, the Caribbean and Ghana - including flare-off gas from oil production that used to be burned off, with geothermal in the Nordics. The US chose not to invest in its grid for 40-50 years and it is decrepit compared with China, so federal money to improve the grid and uniform siting rules across localities would be strongly positive for the build-out.
Nuclear is right long-term data-center power.
Nuclear, including small modular reactors, is obviously the right thing to do for powering AI data centers: over a 20-30 year horizon it is vastly more efficient than any other form of power generation, its only disadvantage is a higher upfront cost with the benefit spread over years, and if he had the ability to build a nuclear-powered data center he would do it for sure. More aggressive nations like the UAE are already building modular-nuclear-based data centers and putting large nuclear capacity on the grid, and Japan is adding new reactors after Fukushima; nuclear is safe and can be made safer. The caveat is timing - it is probably not the power source for data centers over the next three to four years.
AI compute demand still very early.
The AI compute build-out is not overbuilt - demand is still at the very beginning, so no digestion period is needed. Only a tiny portion of enterprises have adopted AI in a way that meaningfully changed their workflows, the most frequent consumer users query models only six to eight times a day and will go to 100 times a day plus their devices acting for them, and every engineer will use a coding co-pilot. Agentic workloads like deep research kick off cascades of queries where each result feeds further queries, so compute per use is rising at the same time as models get better and more people use them more often; inference demand is going through the roof and the next workload keeps arriving.
Memory prices stay high 18 months.
The memory shortage will persist and prices will stay high for roughly 18 months. Buyers got scared when memory makers could not tell them when supply would be available, so they went from placing six months of demand to 18 months of demand; the order signal into the makers exploded even though the industry is making exactly the same amount of memory as four months ago, and it will take about 18 months to digest that signal. This is a known phenomenon in the memory market that happens every six to eight years, but what is different now is that GPUs are chewing through huge amounts of HBM (a flavor of DRAM), leaving less supply for other devices and consumer products.
Subsidized BYD will undercut European automakers.
Heavily subsidized Chinese EVs are set to undercut European automakers. BYD cars are already everywhere across the developing world and in Mexico City, are about to be shipped to Canada, and are getting footholds in Europe; there is no way the Chinese government is not subsidizing them by 30-50%, and it subsidizes at the finished-product level so the entire supply chain - battery makers, transmission makers - benefits. They have been at the English car manufacturers for a while and the Germans, who still make great cars, are next: few buyers will pay $40-60K for a BMW, Volvo or Audi when a comparable BYD costs $20-40K, and tariffs are the main protection against this price dumping.
Subsidized BYD will undercut European automakers.
Heavily subsidized Chinese EVs are set to undercut European automakers. BYD cars are already everywhere across the developing world and in Mexico City, are about to be shipped to Canada, and are getting footholds in Europe; there is no way the Chinese government is not subsidizing them by 30-50%, and it subsidizes at the finished-product level so the entire supply chain - battery makers, transmission makers - benefits. They have been at the English car manufacturers for a while and the Germans, who still make great cars, are next: few buyers will pay $40-60K for a BMW, Volvo or Audi when a comparable BYD costs $20-40K, and tariffs are the main protection against this price dumping.
Up Next
This All-In Podcast video, published January 23, 2026, features Brian Armstrong, Andrew Feldman discussing Cryptocurrencies, STABLECOINS, UNG, Power infrastructure, URA, AI Semiconductors, AI Data Centers, Memory stocks, BMW, VOLVF, European automakers, 1211.HK. 8 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Brian Armstrong, Andrew Feldman · Tickers: Cryptocurrencies, STABLECOINS, UNG, Power infrastructure, URA, AI Semiconductors, AI Data Centers, Memory stocks, BMW, VOLVF, European automakers, 1211.HK