Block's Beat Was Real. The Bitcoin Margin Story Behind the AI Headlines Is Not.
Two days ago, Block Inc. dropped its quarterly stack and the market did the thing it always does during chop: cheered the headline beat, ignored the ledger underneath. Cash App and Square drove better-than-expected results. The company told analysts it has expanded the use of AI across software engineering. The stock popped. The press release got copy-pasted as a victory lap. I read the actual filings instead. The alpha is not in the earnings-per-share line. It never is.
I have been chasing the alpha while the market sleeps since the EOS days, and the rhythm is always the same. A consumer fintech beats consensus on two products, tacks on a shiny AI sentence, and suddenly the whole report is about transformation. The hard numbers tell a different story. The machine generating the most dollar volume on Block's platform is still the bitcoin brokerage buried inside Cash App. That machine's gross profit margin is thinner than a stablecoin pool's spread in a sideways market. The AI announcement changes the cost curve, not the revenue story โ and the distance between those two statements is where the real trade lives.
Let me be direct about my bias. I run a crypto news desk out of Frankfurt. I have watched earnings reports like this one since before the 2017 EOS mainnet theater. Speed over precision when the chart breaks is my rule, but in earnings season precision is the trade. So here is the precise number nobody flagged: the difference between bitcoin revenue and bitcoin gross profit is enormous, structurally, and growing wider with every regression in volatility. The market sees a line labeled "Bitcoin Revenue: Billions" and assumes a crypto bull engine. The math says otherwise.
Context: Why this is a crypto story before it's a payments story
Block Inc. is the rebranded Square. Jack Dorsey's company started as a white plastic dongle that turned iPhones into credit card terminals for small merchants. That business, now called Square, the seller ecosystem, processes card payments for independent retailers. It is boring, heavily regulated, and reliably cash-generative. Cash App is the opposite: a peer-to-peer wallet that added bitcoin purchase and sale in early 2018, then lightning withdrawals, then self-custody momentum through the Bitkey hardware wallet project. That compounding feature set turned a payments app into one of the largest retail bitcoin on-ramps in the United States. Under the hood, the TBD division and Spiral continue to fund open-source bitcoin infrastructure in a way that barely shows up on the quarterly P&L.
Investors keep misreading the relationship between revenue and gross profit. Bitcoin is a pass-through business. When a Cash App user buys $100 worth of BTC, Block recognizes close to $100 in bitcoin revenue and simultaneously pays its liquidity provider nearly the same amount. The gross profit left behind is the spread on those orders, plus a few fee lines. Historically, bitcoin revenue has swung between roughly 30 and 50 percent of total net revenue, while bitcoin gross profit has landed near 2 to 4 percent of total gross profit. You can look that up in any of the company's public financial statements. The disconnect is the entire mispricing story.
The company has added a second narrative layer in this reporting cycle: AI in engineering. Executives said software engineers across the stack now rely on AI-assisted tooling for code generation, test writing, code review, and incident response. That is not a product announcement. That is a cost-structure announcement. And in my audit experience, when a company announces AI inside the engineering organization during the same quarter that two consumer products beat consensus, the announcement is usually doing one narrow job: shifting attention from the thin margin line to the growth line.
Core: What the data actually says
Now the real work. I pulled the clean numbers out of the earnings release and the supplementary data file. Cash App gross profit grew at a high-teen to low-twenties percentage clip on a year-over-year basis. Square gross payment volume came in ahead of the street estimate. Subscription and services revenue, the category that wraps bitcoin gross profit, lightning fees, and buy-now-pay-later products, grew faster than transaction-based revenue. That is a healthy mix in the abstract. Transaction revenue is the stable spine. Subscription and services revenue is the muscle. The trouble starts when you dissect the muscle.
The bitcoin sub-segment demands a closer look. I rebuilt the historical bitcoin gross profit per monthly active user from the company's public data. The pattern is brutally consistent. During the retail frenzy of late 2021, gross profit per bitcoin user spiked alongside price volatility and rising order size. During the 2022 through 2024 sideways grind, that per-user profitability collapsed, even as the count of monthly active bitcoin buyers kept crawling upward. The lesson is straightforward: Cash App's bitcoin economics are volatility-dependent, not volume-dependent. It is an options business wearing a brokerage costume. A dollar-cost-averaging customer who buys $20 a week generates a fraction of the gross profit of a panic buyer who sweeps $5,000 into the market in one tap during a liquidation cascade.
This is where the AI announcement turns interesting. If Block has actually rolled LLM-assisted tooling across its engineering org, the payoff is not faster consumer feature shipping. The payoff is reduced headcount pressure and lower incident remediation cost. I have watched this exact wave inside crypto startups since 2023. When a protocol team adopts AI-assisted code review, its heaviest cost center โ the payroll of senior solidity and backend engineers โ stops growing. The savings are real. The effect on revenue is indirect and delayed. Block is telling the market it can run a larger platform with a flatter engineering payroll, and that flows to operating margin three to six quarters later. But none of that flows to bitcoin gross profit.
My cross-reference with on-chain data sharpens the point. I matched Cash App's reported bitcoin gross profit against a rolling 30-day average of retail spot exchange netflows sampled across major US venues. In high-volatility regimes, the correlation between Cash App bitcoin profit and retail exchange inflows tightens to a dangerous degree. In quiet chop, the correlation breaks apart. That divergence is the entire 2025 story. Retail users keep buying small amounts of bitcoin through Cash App because the interface makes the habit frictionless, but they are not panic-buying. They are stacking sats the way earlier generations bought index funds. Gross profit per order keeps shrinking because order sizes are shrinking and volatility is flattening the spread.
Compare that model with Coinbase. When Coinbase reports a trading-revenue beat, the market instantly prices in volatility and churn. Block gets no such adjustment. Because bitcoin revenue is a pass-through line, the volatility that drives its crypto profit is laundered into a boring services category. That optical honesty creates a real mispricing: the AI narrative lifts the multiple while the ignored volatility lever silently determines the actual bitcoin gross profit.
Add the AI layer and you get a compounding effect with two edges. On the cost side, the platform gets cheaper to operate. On the revenue side, the crypto product morphs into a dollar-cost-averaging utility rather than a trading desk. That is a better business than a pure brokerage, but it is a worse crypto narrative. The market has not yet learned to tell the difference.
Let me go one level deeper into the engineering claims, because I have spent the past year benchmarking AI-assisted development inside crypto companies for my own edge. The standard metrics are code acceptance rate, pull-request cycle time, and the share of new code contributed by AI. A well-run crypto org with LLM-based review tools typically moves its pull-request cycle time down by a third within two quarters, and roughly 20 to 30 percent of newly merged code starts as AI-generated snippets. The danger is the accumulation of invisible technical debt. I have seen teams celebrate productivity gains while their critical dependencies silently rot, because AI test generation is excellent at generating tests that confirm existing assumptions and terrible at reasoning about adversarial edge cases. For a company operating a licensed bitcoin broker, the adversarial edge cases are regulatory, not just technical.
Look at the hiring pattern instead of the press release. Block's public job board is full of AI infrastructure roles: LLM evaluation engineers, machine-learning platform specialists, developer productivity leads. Every one of those roles sits on the cost side of the income statement. Almost none of them touch the bitcoin product directly. That mismatch is the quiet tell. Block is spending engineering capacity on the rails, not on the crypto product. The bitcoin business is running on autopilot as a utility, while the AI story collects the headlines and the multiple expansion.
From the sprint to the sprawl of DeFi, I have seen the same script run at a dozen layer-2 teams. In late 2023, several L2 projects announced "AI-optimized sequencers" with great fanfare. The only measurable effect was a marginally lower error rate in a price oracle. No user-facing adoption shifted. At corporate scale inside Block, the dynamic repeats: AI expands the infrastructure, but the cash register rings for exactly the same reasons it rang a year ago.
Now bring in the regulatory layer, because this report is unreadable without it. With MiCA fully live across the EU and the US still wrestling with stablecoin and market-structure legislation, the compliance burden for a licensed bitcoin broker is not static. Block has spent years assembling licensing infrastructure for Cash App bitcoin: Money Transmitter Licenses in dozens of states, the New York BitLicense, anti-money-laundering controls, travel-rule compliance systems, and tax-reporting floors. Every new regulatory regime adds fixed cost to a product line with a structurally low gross margin. This is where the AI announcement secretly earns its keep. The CFO can point to AI-driven engineering efficiency at exactly the moment the CFO needs an offset against climbing compliance cost. The real marriage of the two headlines is not growth. It is cost substitution. AI savings get redeployed into regulatory overhead, not into bitcoin revenue growth.
I mapped that substitution curve for my own research. Take the known cost structure of a mid-tier US money transmitter, layer on the federal and state filing obligations, and compare the trend line with the reported gross margin of the bitcoin segment. The two lines converge. The company is effectively running in place: every efficiency gain from AI engineering is being absorbed by the widening compliance moat. That is a defensible strategy. It is not the story the market thinks it just approved.
Let me add one more data point from my own crisis playbook. When FTX collapsed in November 2022, I skipped the press releases and traced $600 million in USDC from FTX wallets to Alameda addresses within hours, mapping the capital flight before the exchanges froze withdrawals. That experience taught me to trust flows over filings. Applying that same habit to Block's quarter, the flow that matters is the bitcoin gross profit per user line, not the total revenue line. Total revenue is noise when the pass-through component is huge. Per-user gross profit is signal. And the signal slope is negative in a sideways market. That is the entire thesis in one sentence.
Contrarian: The AI headline is the trap
Here is the angle nobody is reporting. Tracing the EOS endgame back to its genesis block, you find the identical structure: a massive war chest, an infrastructure-first pitch, a chorus of believers, and a slow bleed when usage fails to follow the promise. EOS raised roughly $4 billion through a year-long token sale. It bought block producers, designed an "operating system" for dApps, and spent millions on marketing. The infrastructure was real and the usage was not. When the hype cycle rolled over, price action followed usage, not promises. The EOS lesson is not that the developers were frauds. It is that infrastructure spending without a matching adoption curve is a slow bleed.
Block's AI expansion is infrastructure spending of a different flavor. It is a cost-side efficiency program, and the market is interpreting it as a growth-side signal. That mispricing is the trade. The stock absorbs an "AI lift" while the underlying gross profit composition remains unchanged: payments margins hold, bitcoin margins compress, regulatory costs climb. The AI announcement does not change the intrinsic valuation. It changes the narrative multiple. And narrative multiples always revert.
I will be contrarian about the crypto side too. The most overlooked line item in the entire report is the evolution of bitcoin gross profit margin during a calm market. Reading the room in the order book silence, I notice nobody flags this. In a sideways market, a retail brokerage's per-unit profit falls while user acquisition costs stay flat or rise. The only levers to pull are raising the spread or expanding into derivatives. Block has publicly signaled wallet self-custody and hardware security products, not leveraged trading. So the bitcoin profit line is capped by product choice. The AI story does nothing to lift that cap.
The most dangerous interpretation hiding in this report is the idea that "AI in engineering" means "AI in trading." Nothing in the filings supports that. Block is not an algorithmic trading firm. It is a payments rail with a bitcoin utility attached. The moment a generalist fund manager reads the AI headline and buys the stock as a crypto-AI crossover play, that manager is purchasing a narrative the company never actually sold. Meanwhile, the true signal โ a structurally declining per-user bitcoin gross profit โ stays unpriced until the next volatility shock makes it obvious.
Takeaway: What to watch next
The next earnings print needs to show one of two things: bitcoin gross profit per monthly active user stabilizing in a sideways market, or operating expense as a percentage of gross profit falling faster than guidance implies. If both happen, the AI story is real and the bitcoin margin story has bottomed. If only the AI cost savings show, the squeeze continues under a quieter surface. The chart that matters is not the stock chart. It is the quarter-over-quarter spread between bitcoin gross profit and regulatory cost growth โ until the next volatility regime resets the market.
I have been through these cycles since the genesis-block era. Chasing the alpha while the market sleeps is how you survive them. But the endgame here is not the price of Block stock. It is whether Cash App's bitcoin rails become an irreplaceable on-ramp before the AI cost savings run out. Speed over precision when the chart breaks โ but before the break, precision is the entire game. The order book is listening.
