Spring 2026 has one thread running through banking, technology, and macro coverage. The thread is balance sheets. Hyperscalers are spending unprecedented sums on AI infrastructure. Banks are working out where the next basis point of margin will come from. Stablecoin issuers are quietly accumulating Treasury bills that rival the size of midsize money market funds. These look like separate stories. They are the same story told from different desks.

The bank margin question

Net interest margin is the simplest scoreboard for a bank. Take what the bank earns on assets, subtract what it pays on liabilities, divide by average earning assets. For most of the last two years, large US and European banks earned a friendly spread because policy rates rose faster than deposit costs. That tailwind is fading.

Deposit betas, meaning the share of policy rate changes that pass through to deposit rates, have been climbing as savers shop around. Online challengers and brokered money market funds keep raising the floor. Central banks in Frankfurt and London have started easing in measured steps. The Fed has signaled patience but futures markets keep pricing in a glide path lower.

The result is that the easy phase of bank earnings is over. The next year of results will favor banks that book loans at decent spreads and avoid surprises in commercial real estate. Banks with strong trading and investment banking franchises will look better in the league tables. Regional players that rely on branch deposits will find it harder to keep loan growth profitable.

Where the AI money is coming from

The capex numbers from the largest cloud providers in the United States now resemble a small national budget. Combined annual spending on data centers, accelerators, networking, and power is on track to clear half a trillion dollars before the end of the decade. The official line is that all of it is funded from operating cash flow. That was true in 2023. It is starting to bend.

A handful of names have raised debt to back data center joint ventures. Private credit funds have stepped in for the smaller builders. Some hyperscalers are signing power purchase agreements that effectively pre commit decades of utility cash flows. The financing channel for AI has quietly become a multi layer stack, with public equity at the top, investment grade debt in the middle, and private credit, project finance, and lease structures at the base.

For banks this matters in two ways. They earn fees arranging the debt. They also fund the warehouse lines that private credit funds draw on. When a finance ministry official talks about AI as a strategic asset class, what they often mean is that the loan books behind it are growing faster than anyone modeled.

Stablecoins and the deposit drift

A stablecoin is, in plain terms, a promise to pay one dollar backed by something that should be worth one dollar. The largest issuers now hold tens of billions in short term US Treasuries and overnight repo. Those holdings sit outside the regulated banking system. They earn the issuer the carry. They earn the holder nothing.

This is a deposit drift in disguise. Every dollar parked in a stablecoin is a dollar that did not stay in a checking account. For now the totals are small relative to total deposits. The trend line is not. Regulators in the US, the UK, the EU, and Singapore are all moving on issuer licensing, reserve disclosure, and audit requirements. The shape of the final rules will decide whether stablecoins remain a crypto sidecar or become a real settlement rail for retail and corporate payments.

If they become a settlement rail, the consequences for bank funding are obvious. If they do not, the issuers will still have built a parallel money market fund industry inside a wallet app.

Private credit picking up what banks dropped

The private credit story is not new but the scale is. Direct lending funds and business development companies now hold loan portfolios that look, from a distance, like the leveraged loan books that regional banks used to carry. The asset class has grown through a benign credit cycle. It has not been tested through a sharp downturn at this size.

What is new in 2026 is the convergence of three trends. AI capex needs financing. Banks face higher capital charges on certain corporate loans under the latest Basel implementations. Private credit funds have the dry powder and the regulatory flexibility to fill the gap. The handoff is well underway.

For depositors and shareholders it pays to ask where the credit risk actually sits. A bank that books fewer loans is not necessarily a safer bank. It may simply be a bank that earns less and hands the risk to a vehicle whose investors include the same pension funds that ultimately back the depositors.

What to watch

Three signals are worth tracking through the rest of the year.

First, deposit costs at the large universal banks. If betas keep climbing while loan growth stalls, net interest income will compress and cost cutting will return as a theme on earnings calls.

Second, the disclosed debt issuance and lease structures of the top five US hyperscalers. The shift from pure cash funded capex to mixed funded capex is the cleanest indicator that the AI cycle is moving into a different financial phase.

Third, the supervisory letters and consultation papers on stablecoin reserves and bank deposit insurance. Quiet rulemaking now will decide who holds the next ten trillion dollars of US dollar liquidity.

None of these is a headline by itself. Together they describe the financial plumbing being rewired in real time.