A2Z Fintech #38 by AMWB: Predictions, Podium & Pie.

Monday morning in Singapore, last week. I was back at my desk for the first time in ten weeks, jet lag still settling, coffee going cold, inbox in the kind of dishevelled state that three continents and two conferences tend to produce. WhatsApp buzzes. It is Zubin, calling from the Bay Area with the particular equanimity of a man who has been waiting for this exact conversation since Christmas.

“You can’t just swan back onto this podcast after three months of globetrotting and pretend the scorecard doesn’t exist. The listeners have receipts. I have receipts. Ben and David from Acquired probably have receipts.”

He is right, of course. On the A2Z Fintech Season One finale in December, Zubin and I spent the better part of an hour making ten predictions for 2026. Most people who do that on a podcast quietly hope you forget. We are the idiots who wrote them down, marked them up, and invited the internet to grade us. Ninety days in, the ledger is open. And the only thing that separates a non-obvious insight from a lucky guess is the discipline of being graded in public.

In Formula One, the trophies get sprayed on the podium. The truth gets spoken in the debrief. An hour after the chequered flag, in a windowless room somewhere behind the paddock, a team principal hands the driver a printout. Every corner. Every braking point. Every tenth of a second that went missing. The podium celebrates. The debrief diagnoses. This week, A2Z Fintech did its debrief.

Last week’s blog, Bluffs, Bridges & Bundles, walked through Mastercard’s $1.8 billion BVNK acquisition in real time: the largest stablecoin deal in history, the Stripe-Bridge prequel, and the great unbundling of the card networks. That deal is also the centrepiece of today’s scorecard, because one of our December predictions landed on precisely the day we said it would. This week, we go back to the source: the Christmas edition, A Man Who Blogs #21: The Grand Prix Season Finale, where the ten bets were placed. The full back-and-forth, with all the self-flagellation and the occasional vindication, is in this week’s A2Z Fintech podcast, and this post is the companion piece.

The headline: four out of ten fully playing out, three directionally correct, one spectacularly wrong, and one where the outcome is, shall we say, pending. Between two people with fifty combined years of scar tissue across banking, big tech, and payments, I will quietly submit that is not a bad hit rate for ninety days. But let us not grade on the curve.

Dead right, and on the day. On 17 March 2026, Mastercard announced its $1.8 billion acquisition of BVNK, the British stablecoin infrastructure company. Largest stablecoin deal in history. Visa did not just watch either; Visa Direct is now running at $3.5 billion in annualised stablecoin settlement volume as of the Q1 2026 reporting period. Not a pilot. Not a press release. A product line. And the non-obvious bit the market has not fully priced in: the card schemes are not becoming stablecoin companies; they are becoming the regulated settlement layer for stablecoins. The new interchange. Same toll roads, different asphalt.

Not close. Gartner’s Forecast Analysis: Sovereign Cloud and the Global Shift to Digital Autonomy, 2024-2028, puts sovereign cloud spending at $80 billion in 2026, up 35% year-on-year. The UAE’s G42 announced the Phase II completion of its Stargate-class cluster in February. The UK’s Department for Science, Innovation and Technology confirmed Milestone 1 of its Sovereign AI Unit in January. India’s MeitY gazetted the AIRAWAT-2 charter in March. The Saudi PIF’s Vision 2030 tech supplement committed what reads as 100,000-plus H100/H200-equivalent units of compute capacity. The word sovereign is now a standard prefix to AI, which is the bureaucratic equivalent of a five-star general admitting that the campaign matters. Compute is the new oil. Sovereignty is the new OPEC.

Messy, public, in progress. In late March, the two companies amended the partnership. Microsoft is locked in a reported $250 billion in contracted Azure purchases over the next decade, which sounds like a win until you read the fine print. The restructuring grants OpenAI the flexibility to use third-party compute, including its own in-house Tigris chips, in exchange for Microsoft securing long-term preferential licensing over OpenAI’s Agentic AGI frameworks. Translation: Microsoft has bought optionality on the next platform; OpenAI has bought freedom from the last one. The word divorce is doing work. This is not a clean split. It is a conscious uncoupling with a $250 billion prenup.

The paperweight call. In March, the first LoveFrom/OpenAI prototype leaked. Early hands-on reports describe a tactile, screenless pendant built around haptic feedback and ambient voice — characterised by tech critics as a luxury object where the minimalist aesthetic is doing more work than the utility. I say this as a lifelong Jony Ive fan, a fanboy before the iPhone was launched. But the market is sceptical for the exact reason we flagged in December. Melding hardware and software into a magical object is infinitely harder than Steve Jobs made it look. And Sam Altman, for all his talent, is no Steve Jobs.

Takeaway: The four calls that landed in ninety days shared one thing. Each one read the structural logic of the system rather than the headline. Incumbents co-opt the disruption they cannot beat. Compute is treated like oil the moment states realise it is scarce. Partnerships restructure when the incentives diverge. Hardware is hard. Write those down the week they matter, and ninety days later the press releases start to arrive on schedule.

Halfway right, halfway wrong, fully interesting. There is no signed Google-Huawei joint venture; geopolitics made that one a bridge too far, as it probably always was. What actually happened: In March, Huawei launched the Ascend 950PR processor and the Atlas 350 server, with internal benchmarks claiming 2.87x performance over Nvidia’s H20 in LLM training and inference throughput. The important detail, the one that strengthens the prediction rather than weakens it: the H20 is the throttled China-export model, not Nvidia’s flagship H100 or B200. Which is precisely the point. Huawei is winning the race Washington has permitted them to run, and building the ecosystem to compete in the races Washington has not. Meanwhile, Google is aggressively pushing its OpenXLA stack to decouple software from Nvidia’s hardware, and Huawei is building to the same open standards. The alliance is real. It just is not a press release. It is an open-standard handshake, which in a geopolitical cold war is arguably a more durable form of partnership than shared equity. In a world where American export controls make a formal JV impossible, the new alliances form through shared standards, not shared balance sheets. A lesson I will take from the miss.

Directionally correct. Wrong in the specifics, and the specifics matter. We predicted regulators would create a dedicated license for AI agents to hold money and sign contracts. What actually happened is more consequential, and it happened in three places at once.

In March, the UK’s Competition and Markets Authority published its Guidance on Agentic Liability and Algorithmic Consumer Protection, establishing the deploying corporation as the ultimate liable entity. Singapore’s MAS released its Autonomous Agent Commercial Framework consultation paper for financial institutions. Brussels folded Annex IX into the 2026 AI Act update, introducing agent-licensing requirements for autonomous systems managing assets over €1 million.

Three jurisdictions, three different instruments, one direction of travel. Different from a license. Arguably more potent. If your agent goes off the rails and books ten thousand flights to Kazakhstan on a corporate card, you cannot hide behind the AI did that. The legal framework is being built faster than we expected, through liability and threshold-licensing rather than blanket registration. That distinction is going to shape which industries adopt agents in Q2 and Q3, and which keep them in the sandbox.

No Q-Day event. But the defensive upgrade cycle we predicted is completely real. NIST’s post-quantum standards now sit inside every major bank’s three-year roadmap. The Bank of England issued migration guidance in February. JPMorgan has integrated ML-KEM (Kyber-based) encryption into its cross-border settlement rails. Standard Chartered has moved a quantum-resistant VPN and identity-management pilot into production in its Hong Kong and Singapore hubs. And research published in Q1, building on the Schnorr/Regev refinements with shallow-circuit logic and advanced error correction, has dropped the physical qubit requirement to crack RSA-2048 from roughly 20 million to between 850,000 and 1.2 million — a factor-of-twenty reduction in the security margin. The actual break is closer than the industry wants to admit. The scare is here. The event is not. Half credit and a nervous glance at the calendar.

Takeaway: The three close calls share a pattern. We read the direction of travel correctly and mis-specified the mechanism. Alliances, liability regimes, and cryptographic risk all arrived on the timeline we flagged; they just arrived wearing slightly different clothes. Hold the direction with conviction and the mechanism loosely. The river moves; the shape of the channel is found along the way.

Dead wrong. No movement. Jamie Dimon’s Q1 commentary was all about sovereign AI infrastructure. Nubank remains independent, growing, US-listed, and frankly looking more like the acquirer than the acquired. David Vélez is not waiting to be bought; he is preparing to invade the United States, a theme we covered in AMWB #27: Purple Power & Persistence. If anyone is the hunter now, it is him.

One small consolation. Our corollary, the afterthought we almost cut from the script, has aged beautifully. We said the JPM move would trigger reverse mergers where large neobanks would buy small traditional banks for the license and the deposits, not for the customer book. That is exactly what has happened. In Q1, Revolut acquired a distressed regional lender in Germany to secure a full domestic deposit license; Monzo executed a licence-grab of a small Nordic commercial bank; Starling picked up a French mortgage book and licence. Three deals, three geographies, one thesis. I am being defensive. Zubin will call me out on it.

This was mine. I said Ben and David at Acquired.fm would make us a Standard Oil offer to join their empire. I said we would politely decline and stay independent. Power to the people.

Reader, they did not call. Not a hostile bid. Not a friendly expression of interest. Not a passing reference on their own podcast. Nothing. I would like to blame this on tightening M&A credit spreads, a backchannel rebuff we strategically buried, or a clerical error at their end. I cannot. The unvarnished truth is that we manufactured a takeover offer that nobody was offering, declined it on air, and walked off the stage as though we had dodged a bullet that was in fact a very small pebble thrown by nobody.

Zubin has been generous about this. He also points out, correctly, that the right F1 analogy is not Max Verstappen rejecting an offer from Mercedes. It is Max Verstappen announcing, unsolicited, that he is declining an offer from a Formula Three team that does not exist.

Takeaway: The fails teach more than the wins. The JPM-Nubank miss was a failure of method: we mistook symmetry for structure. The Acquired.fm miss was a failure of scale: we mistook our own importance for the market’s interest. Different errors, same diagnostic value. The most dangerous prediction you can make in public is the one about your own importance.

Anyone can publish a scorecard. The question worth sitting with, after Zubin and I closed the tally, is the one underneath. Where did the method work, and where did it fail?

The signal, when I read the hits and misses together, is legible. We got it right where we read the structural logic of the system. Stablecoins displacing card rails was always going to be a capture, not a war; Visa and Mastercard do not lose to disruption, they absorb it, and the only questions were price and timing. The OpenAI–Microsoft tension was always going to end in a divorce, because the incentives had diverged and no amount of messaging discipline can paper over a structural conflict of interest. Compute nationalism was inevitable the moment GPUs became strategic assets; states do not watch scarce strategic assets flow through private borders for very long. These were calls based on fifty years of watching how large institutions actually behave under pressure, not on reading the headlines.

We got it wrong where we substituted the narrative for the structure. The JPM–Nubank merger was a beautiful story. Jamie Dimon finally gets his global retail bank. David Vélez cashes out at a premium. The symmetry was irresistible. But symmetry is a storytelling instinct, not an investing instinct, and the system was writing a different script.

Call that failure mode The Narrative Trap: the predictive error of betting on the story that resolves cleanly instead of the system that actually runs. Every board, every executive team, every investor is susceptible to it. So are we, and we write this down for a living.

The counter-discipline is less glamorous. Looking around corners is not about being psychic. It is about knowing which parts of the system are load-bearing and which parts are just decoration. Incentives are load-bearing. Cost structures are load-bearing. Regulatory architecture is load-bearing. Founder equity is load-bearing. Headlines, narratives, M&A speculation, the urge for a tidy ending: decoration.

Takeaway: The Narrative Trap is not a failure of research. It is a failure of discipline. The antidote is the scoreboard itself: written down, dated, graded in public. When we read the load-bearing walls, we got it right. When we read the decor, we got it wrong. If there is a single transferable lesson from this quarter’s ledger, it is that one.

If Q1 has a single signal, it is that the Mastercard-BVNK deal was a starting gun, not a finish line. Four new predictions for the rest of 2026, three mine and one Zubin’s, with sharper stakes and shorter clocks.

Prediction #1 (Aman): Visa acquires Rain or Zerohash before Q3 closes. Visa cannot rent the infrastructure that Mastercard now owns. Every week of hesitation narrows the target list and widens the multiple.

Prediction #2 (Aman): Circle acquires or merges with a payments distribution company, like Checkout.com. Being the issuer of the second-largest stablecoin with no distribution leverage is an increasingly untenable position.

Prediction #3 (Aman): The next wave of reverse mergers arrives, and they get larger. By Q4, I expect to see at least one transaction in the $500 million-plus range, most likely a neobank acquiring a mid-tier licensed bank in the US for the charter.

Prediction #4 (Zubin): Watch the BVNK integration window. If Mastercard has not shipped a live, merchant-facing product on BVNK rails within nine months, Visa catches them flat-footed.

Every quarter, boards and executive teams ask some version of the same question. Where do we sit in all of this? Are we the architect, the plumber, or the tenant?

That is the work we do at A2Z Advisors. Strategic clarity sessions, keynote provocations, and deeper boardroom engagements. The goal is always the same: leave the room with a sharper view of where the next ten years are heading and where you are actually positioned within them. The same fifty years of collective scar tissue that let us see the Mastercard move in December, and miss the Nubank one, is the same fifty years we bring to the conversation. The scoreboard discipline travels.

If you lead a bank, a fintech, a payments platform, or an investor navigating the great unbundling, the agent economy, or the stablecoin arms race, drop me a line. First conversations are always free. Second conversations are almost always worth having. I am in Singapore through April, London and Dubai in May, and the Q3 diary is filling faster than usual.

The original predictions episode, The Grand Prix of Tech Finale, is where all of this started. The companion blog from last week on the Mastercard-BVNK deal is Bluffs, Bridges & Bundles.

We will see you at the half-season review in July. By then, some of the directionally-correct calls may have graduated to dead right. Some of the dead right may have wobbled. That is the beauty of making real predictions in real time in public. The scoreboard never lies.

Until next week.

Stay curious.