I don’t usually write about crypto in this newsletter — my beat is practical AI for business leaders. But this summer, the two worlds collided hard enough that I spent the last two months tracking it closely, reading through roughly forty newsletter issues on the sector. What I found wasn’t speculative mania.
It was financial infrastructure being rebuilt in public, with all the growing pains that implies: layoffs at crypto’s most important nonprofit, a major hardware-security failure, regulators suing each other, and — most relevant to us — AI agents quietly getting their own wallets and spending limits at every major exchange. Here’s the condensed picture, organised by theme.
Ethereum is rebuilding itself while reorganising.
Ethereum spent the summer doing two things at once: sketching an ambitious technical future and going through real organisational upheaval. On the technical side, researchers published proposals for what’s being called Ethereum’s “third major iteration” — nodes verifying compressed proofs instead of re-executing every transaction, plus a “fast finality” mechanism that cut confirmation time from roughly 15 minutes down to a single block in year-long testing.
If it converges as planned, by late 2027 Ethereum’s Layer-2 networks and its base layer become one unified system.
Organisationally, it’s messier. The Ethereum Foundation cut about a fifth of its staff in late June and reorganised into five focused teams, while several of its most senior researchers left to start independent, well-funded spinouts — Ethlabs for core protocol research, EthSystems for institutional privacy tools, and Ethereum Institutional as a dedicated liaison to Wall Street.
Read together, it looks less like a retreat and more like Ethereum’s centre of gravity shifting from one foundation to a constellation of specialised organisations. If you’re advising anyone on “official” Ethereum direction, know that the answer is more distributed than it was in January.
Robinhood’s new blockchain got hijacked by memecoins.
Robinhood launched its own blockchain on July 1, built for tokenised stocks and regulated finance. Within 48 hours it had become the summer’s memecoin capital instead — trading volume passed $900 million in its first week, one meme token was briefly up 5,000%, and by month’s end the chain had overtaken Coinbase’s own network in daily users and revenue. The intended use case (trading real stocks onchain) is still there and still growing, but it got completely overshadowed.
The more useful pattern underneath the noise: infrastructure providers are racing to own the “launch” layer directly instead of just supplying the plumbing beneath it. Uniswap, previously just the exchange everyone traded through, built its own token launchpad. Two Solana platforms, Pump.fun and a fast-growing challenger called Fomo, are now in open warfare over the same social-trading layer, with one reportedly paying top traders real money each month just to switch sides.
If you’re watching any platform business for competitive dynamics, this is the same story you already know from social media and app stores: whoever owns distribution and trust wins, not whoever owns the underlying pipes.
Stablecoins: the quiet land grab.
This is probably the most consequential story of the summer, even though it made the fewest headlines. A new stablecoin consortium called Open USD launched with heavyweight backers — Stripe, Visa, Mastercard, Coinbase, BlackRock, Google, Shopify — built on a structure that shares reserve income with the banks and fintechs that distribute it. That’s a direct challenge to Circle’s USDC, which currently keeps most of that revenue for itself.
Circle’s stock fell nearly 20% on the announcement and kept sliding as reports emerged that its contract renegotiation with Coinbase — its biggest partner, and now also an Open USD backer — had lost leverage.
Layered on top: unconfirmed but widely discussed reports that Stripe and a private equity partner offered $53.4 billion for PayPal, which would be the largest fintech acquisition in history and would pair Stripe’s stablecoin infrastructure with PayPal’s enormous consumer base. Whatever happens with that specific deal, it tells you where large payments companies think the strategic value now sits.
The practical takeaway for anyone advising on payments: stablecoin economics are shifting fast, from the companies that issue the currency toward the companies that distribute it. Expect more consortium-style launches like this one.
AI agents are getting wallets — this is the one to watch.
If you take away one section from this post, make it this one. Over the summer, essentially every major crypto exchange and wallet shipped infrastructure specifically for AI agents to hold and spend money autonomously, with guardrails attached.
Coinbase launched isolated sub-accounts with spending limits and a kill switch, plus an SEC-registered AI investment advisor. Robinhood, Kraken, Gemini, and MetaMask shipped comparable features in the same few weeks. Cloudflare — which sits in front of roughly a fifth of the entire web — unveiled programmable stablecoin wallets built specifically for agent-to-agent payments.
The payment rail underneath a lot of this is a protocol called x402, which lets an AI agent pay for an API call or a dataset in stablecoins with zero human involvement.
Amazon integrated it to charge AI traffic at the network edge, and Cloudflare opened a waitlist letting any website charge bots and agents to access its content. Worth noting: Cloudflare’s own CEO admitted current blockchain throughput can’t yet handle this at real scale, so the intent is real but the plumbing isn’t fully there.
A second, unresolved problem is agent identity — how a website tells a legitimate AI agent apart from a scraper. New standards are emerging, but adoption is thin; one study found barely one in ten registered agent identities were actually functional.
For those of us advising businesses on AI adoption, this is the clearest bridge yet between the tools your clients are already using and the financial rails most of them still don’t understand — and it’s moving from theoretical to shipped in a matter of weeks.
Prediction markets are booming — and stuck in court.
Kalshi, Polymarket, and their sportsbook rivals had their biggest quarter yet, with a combined $50 billion-plus traded in July alone, driven largely by World Cup betting. DraftKings, FanDuel, and Fanatics all built regulated prediction-market products to offer sports betting in states where their own sportsbooks are banned, and Fanatics went further by acquiring its own regulated exchange outright.
Growth is running well ahead of legal clarity, though. The incumbent futures exchange CME sued federal regulators over how these products are classified, and several states are suing to shut Kalshi down entirely, arguing it’s using federal status to dodge state gaming law.
Federal regulators have twice invoked emergency powers this summer just to keep Kalshi operating against state shutdown orders. This is genuinely unresolved — the outcome will decide whether prediction markets become a permanent, boring fixture of US finance or stay in a legal grey zone.
Two security wake-up calls.
A firmware flaw in a popular hardware wallet — the random number generator existed in the code but wasn’t actually running — has now been linked to over $100 million stolen from wallets set up on old firmware. It’s a useful reminder for anyone who assumes “hardware wallet” automatically means “secure”: the good ones use multiple independent sources of randomness and are built to fail safely, not just fail quietly.
Separately, the privacy-focused cryptocurrency Zcash had its own scare: a bug in its core cryptography, present since 2022 and caught only this spring via AI-assisted code review, could theoretically have allowed counterfeit coins to be minted undetected.
Their response is a genuine case study in doing security right under pressure — an emergency patch within days, followed by two months of work to produce a fully machine-verified mathematical proof that no counterfeit coins had ever existed, before reopening the system.
Two smaller DeFi lending exploits in the same period, worth about $6 million each, both traced back to the same root cause: complex, multi-layered “automated yield” products that even their own curators hadn’t fully audited. If you have clients evaluating anything marketed as automated or curated yield, that’s the question to ask.
Regulation: slow in Congress, faster at the agencies.
The crypto market-structure bill that the industry has been waiting years for — CLARITY — advanced out of committee in May but has stalled since on an unrelated fight over an ethics provision, one that gained urgency after a presidential financial disclosure reported over $1.4 billion in crypto-related family income last year.
Betting markets put the odds of the bill passing this year at under 40%, down from over 70% in the spring, and it’s now racing an approaching midterm election calendar. If any client needs regulatory certainty before committing to onchain products, the honest answer right now is: not yet, and possibly not this year.
Where Congress has stalled, regulators have moved faster. The SEC’s crypto point person published guidance warning that the more human judgment goes into managing an “automated” investment vault, the more likely it is to be regulated as a security — directly relevant to the yield products mentioned above.
That built on an earlier formal ruling that classified Bitcoin, Ethereum, Solana, and XRP as commodities and laid out clear rules for mining and staking. It’s real, useful clarity, even without the bigger bill.
Institutional money kept arriving — unevenly.
BlackRock launched new tokenised treasury products designed specifically as reserves for stablecoins, and Wall Street’s central clearing utility processed its first live trades in tokenised stocks and bonds with BlackRock, JPMorgan, and Goldman Sachs all participating. That’s the adoption story continuing on schedule.
But it’s not uniformly rosy. Michael Saylor’s Strategy — the company famous for holding Bitcoin as a corporate treasury strategy — came under real financial strain in late June, with its share price falling to a two-year low and analysts estimating it would need to sell billions of dollars of Bitcoin just to restore a comfortable cash cushion for its dividend obligations.
It’s a useful counterweight to keep handy: the “corporate crypto treasury” playbook that got so much attention in 2024 and 2025 has real financial mechanics behind it, and those mechanics are now being tested in public.
DeFi: innovating and consolidating at the same time.
On the innovation side, new lending products are bringing fixed-rate, bond-like structures to a space that’s historically been all floating rates, and new due-diligence tools are finally making it possible to actually audit what’s inside opaque “automated yield” strategies — directly relevant given the exploits above.
On the consolidation side, well-capitalised platforms spent the summer buying up smaller infrastructure and distribution companies, while thinly capitalised ones either got acquired or shut down outright — including one long-running exchange that closed voluntarily after eleven years with a perfect security record, and three companies that filed for bankruptcy. That bifurcation between the well-funded and the rest is likely to keep accelerating.
The bottom line.
Crypto stopped behaving like a speculative sideshow this summer and started behaving like financial infrastructure being built in public — complete with reorganisations, lawsuits, and security failures.
But the thread that should matter most to this audience is the quiet one: AI agents got wallets, spending limits, and payment rails at every major platform within a matter of weeks. That’s not speculation. It’s already shipped. If you’re helping executives think through what AI-driven business models look like in the next two years, this is the infrastructure quietly being built underneath them — worth understanding well before your clients start asking about it.
As always, treat any specific figures, deal terms, or legal outcomes above as a snapshot in time — this is a fast-moving space, and I’d encourage verifying anything time-sensitive before you act on it.