Coinbase CEO's Top 3 Crypto Trends for 2026 + More from Davos!

Why does it matter? Because crypto just got legal legitimacy — and the people it threatens are already trying to kill it.

Brian Armstrong sat down at Davos to lay out where crypto, AI, and global finance are actually heading — not where the hype says they're going. What emerged is a picture of regulatory wins being quietly contested, infrastructure build-outs that are nowhere near their ceiling, and a CEO who's now asking his AI what he should be thinking about rather than the other way around.

• Regulated stablecoins are structurally safer than banks — and bank trade groups are lobbying to undo the law that makes them so
• 4 billion adults globally can't invest in quality assets; on-chain tokenization is the first real mechanism to change that
• AI inference demand is still in its first inning — current heavy users query 6-8 times a day, and the ceiling is 100+ times daily across all devices
• The US chip lead over China is self-reinforcing, but China is already ahead on open-source models and grid infrastructure

The Genius Act made stablecoins safer than banks — so naturally, banks want it dead

Under the Genius Act, US-regulated stablecoins must hold 100% of assets in short-term US Treasuries — 30-day max. Armstrong is direct about what that means: "30-day treasuries are the safest thing you can get. You're basically trusting the United States government is not going to fail in 30 days."

Compare that to fractional reserve lending. Banks lend out your deposits, keep a fraction in reserve, and carry enormous regulatory overhead precisely because a run is always possible. Silicon Valley Bank proved the point — Armstrong was literally in a board meeting when the run started on a Thursday afternoon, insisted they pull half their cash, and the bank collapsed that night.

Five of the top 20 global banks are now building on Coinbase infrastructure. BlackRock and Apollo have publicly committed to tokenizing every product they offer. The institutional on-ramp is open. But four months after the Genius Act passed, Armstrong is watching something concerning: "The bank trade groups, which I believe are trying to undo the Genius Act — for us, that's a red line."

This is the classic incumbent playbook. Some bank CEOs Armstrong met at Davos are fully leaning in — one top-10 global bank CEO told him crypto is his "number one priority" and "existential." Others are working behind the scenes to claw back the regulatory clarity that makes stablecoins a real competitive threat. The rewards structure negotiated into the Genius Act — not interest, technically a rewards program requiring customer activity — is the business model at stake. If it holds, crypto companies and banks can both win. If it gets gutted, the clock resets on a decade of regulatory progress.

4 billion adults can't invest in anything — tokenization is the biggest wealth-access story since the mutual fund

People have heard about the unbanked. Armstrong's team just published research on a different problem: 4 billion adults are "unbrokered" — they have no ability to invest in high-quality assets at all. The only way they generate wealth is through labor.

Coinbase launched a product called Coinbase Tokenize to address exactly this. Any fund, real estate project, or private company can tokenize its products through the platform. The pitch is straightforward: democratized access, increased demand, elimination of back-office fees, instant on-chain settlement. BlackRock and Apollo have already said publicly they want every single product tokenized. It's not a future scenario — it's happening now.

The second-order effects get interesting fast. Onchain capital formation could let a retail fund raise from tens of millions of people in minutes at $100 or $1,000 entry points. Private companies could go public entirely on-chain, skipping the Sarbanes-Oxley gauntlet that turned Uber, Airbnb, and Instacart into multi-year indigestion stories for late private investors. The SPV market is already a boiler room — operators charging civilians 10% load fees with no carry just to get exposure to SpaceX or Anthropic.

The deeper point Armstrong makes is structural: capitalism's wealth engine — owning productive assets — has been gated behind accreditation requirements that function as a regressive tax. On-chain rails change the denominator. The picks-and-shovels play is whoever owns the tokenization infrastructure at the bottom of that stack.

AI inference is a Netflix-to-streaming moment — not faster DVDs, an entirely new medium

Andrew Feldman of Cerebras arrived at Davos carrying a wafer-scale engine chip — 56 times larger than an Nvidia B200, 4 trillion transistors. The company's founding thesis, built six years before ChatGPT launched: build something not 2x or 5x faster, but 20 to 50x faster, and wait for the demand to arrive.

The demand arrived. Feldman's framing of what speed actually does is worth preserving exactly: "When the internet was slow, Netflix delivered DVDs in envelopes. When the internet got fast, Netflix didn't get better at delivering DVDs. Netflix became a movie studio. It enabled them to be something different. It wasn't a change in degree. It was a fundamental change in kind."

Cognition uses Cerebras to power their coding engine with zero latency between request and answer — developers stay in flow state. Paul Graham captured the cost of slowness in a tweet Feldman quotes: "I'd use Google half as much if ChatGPT weren't so slow." While you're waiting for Claude or ChatGPT, you wander somewhere else. The customer is gone.

OpenAI just signed a 750 megawatt deal with Cerebras — announced at Davos — explicitly to exploit this latency gap. The deal is measured in power, not chip units, because power is now the binding constraint on data center scale. Feldman's view on whether we're overbuilding: current heavy consumers query AI 6 to 8 times a day. When that reaches 100 times daily, when every device is querying autonomously, when enterprise adoption moves from "tiny" to standard — the compute demand explodes by orders of magnitude. "The models are getting better, more people are using, they're using more often, and the amount of compute taken with each usage is increasing. We're just at the beginning."

Armstrong's AI doesn't answer prompts — it tells him what he should be thinking about

The most concrete AI-as-advisor example in the episode came from Armstrong's own operating practice. Coinbase built an internal AI model connected to every data source in the company — every Slack message, Google Doc, Salesforce record, Confluence page. Every team runs on it.

Armstrong's use case has evolved past the obvious. He's stopped asking it to write memos. Now he asks: "What should I be aware of in the company that I might not be aware of?" The system flagged a strategic disagreement on one of his teams he hadn't known about — because it could read every Slack thread and document simultaneously.

Toby, a Shopify executive on his board, gave this a name: reverse prompting. Instead of telling the AI what you want, you ask it what you should be thinking more about. The system reviewed how Armstrong actually spent his time last quarter, compared it to how he said he wanted to spend it, and surfaced the gap: "You actually spent 32% of your time on this instead of 20."

This is a qualitative shift in what AI is for inside operating companies. The tool framing — prompt in, output out — is being replaced by something closer to an always-on chief of staff with total organizational memory. The leverage isn't in faster memo drafting. It's in proactive synthesis across the full information environment of a complex organization.

Today's management layoffs are SaaS-driven — the actual AI displacement wave is still coming, and it will hit whole categories at once

Feldman is precise about something most people are getting wrong. The current wave of middle management cuts at Meta, Google, Microsoft, and others? That's not AI. That's the delayed consequence of good SaaS tools expanding what any manager can actually oversee. The role of middle management — moving information, tracking small teams — shrank in value years ago. Zuckerberg and Nadella looked up one day and asked what those layers were actually doing.

"What I think has happened is this is the delayed impact of good SaaS tools. Your ability to extend your reach as a leader, your scope is much much bigger." The org is flattening because spans of control expanded — not because agents replaced anyone.

The real AI displacement, when it comes, will look completely different. "What we're going to see down the road is whole categories that are vastly more efficient and therefore need less people." Not marginal headcount reductions. Categorical elimination of entire job functions.

Misreading the current moment as AI displacement already peaking creates dangerous policy confusion. It also creates false comfort. The shock hasn't happened yet.

China leads on open-source models and grid power — the US chip advantage is real but narrow, and we're already losing allies

The US-China AI race isn't a single contest — it's a stack, and each side is winning at different layers. Feldman's assessment is cold-eyed: within a few square miles of Santa Clara, six of the world's ten best chip teams operate. Intel, AMD, Nvidia, Cerebras, ARM. "The way you get good at building high-speed chips is to build high-speed chips" — and China hasn't had enough iterations to close that gap. Yet.

But China moved faster on open-source models, precisely because top-down economic control let them make a national decision to modernize their grid and direct compute at the problem. "They have pushed ahead in the open model category." On grid infrastructure, the US is behind.

The 5G parallel is the warning. Huawei ran the table in Africa and the developing world against Cisco. "They clobbered us. And now those places have spyware." The same dynamic plays out in AI if allies end up building on Chinese model infrastructure because US chip policy treated the UAE, Saudi Arabia, and Denmark as second-tier partners.

Feldman's policy read: the previous administration made a strategic error by restricting chips from clear allies while conflating ally access with adversary access. The correction — empowering allied nations to build on US infrastructure standards — is the right move. Recursive systems that build on themselves reward early leads. Getting allies inside the US ecosystem means their development compounds on top of ours. Keeping them out hands China a vector that took decades to fix in telecom.

The real stakes: whoever sets the infrastructure standard wins the compounding game

What this episode reveals, underneath the specific claims about stablecoins and inference chips, is a single pattern: the early infrastructure layer in any technological transition determines who compounds fastest over the next decade. Crypto rails, inference chips, tokenization platforms, chip ecosystems — they're all versions of the same bet. The Genius Act fight, the OpenAI-Cerebras deal, the push to supply allies with US chips: each is a contest over who owns the foundation other people build on. The side that wins the foundation wins everything that comes after it.

---

Topics: crypto, stablecoins, Coinbase, AI inference, AI chips, Cerebras, tokenization, private markets, job displacement, US-China AI race, data centers, energy infrastructure, robotics, Gecko Robotics, Davos, regulation