Bbva bets the bank on ai while rivals still read the manual

War in Europe, energy rationing, tokenised collateral, DeFi eating the margins—BBVA’s answer to all of it is a 127 000-person algorithm. The bank told shareholders on Friday that artificial intelligence is no longer a pilot; it is the strategy, the balance sheet and, if the numbers hold, the next decade’s profit engine.

A record year before the code kicked in

Carlos Torres Vila opened the 2026 AGM in Bilbao by reminding the room that BBVA already pulled off one wholesale reinvention when peers were still installing fax machines. The boast is measurable: 11,5 million new customers last year, two-thirds captured through a screen, and an all-time-high €10,5 billion bottom line that pushed return-on-tangible-equity to 19,3 %. The sub-text: if we digitised once and won, we can do it again—only this time the workforce trains the model, not the other way round.

Onur Genç, the Turkish CEO who keeps a tight leash on cost-to-income ratios, was more blunt. “Our business is people serving people” came the obligatory line, immediately followed by the caveat that those people will soon be augmented or replaced by neural nets. Three battlegrounds are already mapped: client-facing bots that pre-empt a complaint before it is typed; mid-office routines that shrink loan approval from hours to seconds; and a back-office productivity drive that targets a double-digit drop in operating expense by 2028.

The 2028 clock started when altman rang the bell

Genç conceded that the roadmap is “a living document” because Sam Altman’s timeline—super-intelligence landing in roughly 30 months—makes a mockery of five-year banking plans. BBVA’s retort is to lock in vendor-agnostic large language models now, then swap them faster than regulators can spell “model risk”. The bank has quietly moved €400 million of annual tech spend into an ai bucket ring-fenced from the usual quarterly cuts; that line item did not exist in the 2023 budget.

Investors clapped, but only once. The ghost of the failed Sabadell takeover still hovers, and the shareholder register knows that scale matters when the cost of training a frontier model equals the GDP of a midsize country. Torres dodged questions about fresh bids, yet the message was clear: organic growth fuelled by data is cheaper—and politically safer—than another €12 billion cash-and-stock drama.

From pilot to p&l in 18 months or bust

Inside the slide deck shown to analysts, BBVA claims 650 ai use cases live across 8 countries, 40 % of them already hitting “material” thresholds under European banking disclosure rules. The jargon hides a simple bet: every five basis points saved on cost of risk, every 50 basis points gained on fee margin, will be traced back to a model that keeps learning while competitors are still filing ethics paperwork.

The catch? Regulatory heat maps are reddening. The European Central Bank wants explainability; the Spanish data-protection agency wants audits; and the Fed, where BBVA has a toehold through its Mexican arm, wants proof that bias is not being exported across borders. Genç’s team answers with a “glass-box” repository—every model version, every feature weight, every training snapshot stored for a decade. The cost of compliance is ballooning faster than the savings, a tension the bank prefers not to quantify.

Still, the maths is brutal for rivals. A 19 % ROTE in a flat-rate environment means BBVA can outspend most European lenders on R&D without chewing capital. The board has approved a second buy-back programme, €2 billion queued for 2026, funded by the same algorithms that trimmed default rates on consumer loans by 15 basis points last quarter. In short, the market is financing the very experiment that may shrink it.

Bottom line: the race is measured in epochs, not quarters

By the time the next AGM rolls around, BBVA will know whether its data moat is wide enough to keep US platform banks and Chinese fintechs at bay. If the bet misfires, shareholders still leave with a 5 % dividend yield and a cleaner cost base. If it pays off, the phrase “Spanish bank” will sound as quaint as “Spanish empire”. Torres closed the meeting with a reminder that 2025 profits were once considered impossible by City analysts. He did not need to add that impossibility is just another dataset waiting to be retrained.