The orchestration layer
When rails commoditize, the routing decision is where the economics compound, but also where the systemic risk concentrates.
AWS generated $35.6 billion in Q4 2025, a 24% year on year acceleration, and is now running at a $142 billion annualized rate at 35% operating margins. The physical infrastructure underneath that business, servers, storage, networking, is a commodity. Anyone with a credit card can buy the same boxes. What AWS sells, and what the market is paying for, is the layer that sits above the boxes. Identity. Observability. Policy enforcement. Cross region failover. The coordination of thousands of services into something an enterprise can actually operate. The hardware became cheap, and the control plane is what became the trillion dollar business.
I think the same migration is now happening in payments, and most bank strategy conversations that I hear are still framed one layer too low. The rails, FedNow, RTP, ACH, wires, card networks, and now onchain networks, are starting to interoperate and commoditize. The instruments, tokenized deposits and stablecoins, are settling into a two tier structure that banks have already decided they need to support. The value that is compounding, and will compound fast, sits above the rails and the instruments. SWIFT, in announcing the move of its blockchain shared ledger to MVP this year, called it a “shared digital orchestration layer.” JPMorgan describes Kinexys as a control plane for money movement. The terminology is more than marketing I think. It is more a structural claim about where value capture is moving.
I do think that claim is right, with two qualifications. The terminology itself probably deserves more scrutiny than it usually gets. And the cloud computing analogy, which I hear many people using, breaks in a specific place that matters, the place where balance sheet, clearing, and settlement finality live. I think both qualifications change what a bank should actually do about it.
What “control plane” actually means
The phrase is actually borrowed from networking, and the original meaning is pretty precise. In a router, the control plane decides how packets get forwarded, the routing table, the protocols that maintain it, the policies that govern access. The data plane does the actual forwarding. The two are deliberately separated. AWS adopted the same distinction. The control plane is the API surface that lets you create, configure, and manage resources. The data plane is the resource itself doing its job. So even when the control plane fails, the data plane keeps running.
That distinction maps reasonably well onto payments. The rail, RTGS, ACH, FedNow, RTP, a tokenized deposit ledger, a stablecoin network, is the data plane. It moves the value. The orchestration layer is the control plane. It decides which rail, which liquidity pool, which FX path, which compliance wrapper. SWIFT’s own description of its forthcoming ledger reads like a textbook control plane definition: it records and validates interbank payment commitments, provides a synchronized view of obligations, sits above existing rails rather than replacing them.
The terminology fits, but with one caveat the cloud world doesn’t have. In networking, the data plane carries packets, which are pure information. Lose a packet, retry it. The forwarding decision is reversible at almost zero cost. In payments, the data plane carries money. Once a payment is final, it is final. The decision the control plane makes is not routing in the cloud sense, it is a financial commitment, often involving credit, FX risk, and settlement obligation. The control plane in payments doesn’t just route, it implicitly underwrites every routing decision it makes, or it relies on whoever owns the underlying balance sheet to underwrite it. That is what makes this layer harder to commoditize than its cloud equivalent, and what most of the orchestration narrative glosses over.
What the orchestration layer actually does
Strip the abstraction back, and I think the orchestration layer makes four decisions on every cross border or complex domestic transaction. Which rail. Which pool of liquidity. Which FX path. Which compliance wrapper. Each decision has to be made in real time, against the client’s stated outcome, and each has financial and regulatory consequences that don’t unwind cleanly.
Let’s take a corporate treasurer initiating a $50 million payment from a Singapore subsidiary to a German counterparty, due same day. The orchestration layer evaluates correspondent banking through SWIFT, the option of using SWIFT’s forthcoming blockchain ledger or Partior for 24/7 atomic settlement, the option of moving via the bank’s own tokenized deposit rail, the option of converting to a stablecoin and settling on a public chain, and the FX path for each. It applies sanctions, BSA, and credit checks against every leg. It picks the routing that meets the deadline at the lowest total cost including FX spread, fees, and any liquidity funding cost. It executes. It reports the outcome on a single screen, with the ability to override.
This is really not a user interface, nor is it a dashboard. It is a decisioning engine with embedded liquidity access, embedded FX execution, embedded compliance, and the authority to commit balance sheet on the client’s behalf. The screen is downstream of the decisioning. Cross border fintechs figured this out a decade ago, which is why Airwallex is now processing $266 billion in annualized transaction volume at a $1.2 billion ARR, and why Wise moved £181.7 billion across borders in its FY26 with 75% of transactions now settling instantly. Wise Platform, the white labeled version of its infrastructure, is now around 5% of the company’s cross border volume, which is to say banks are paying Wise to do their orchestration for them. Neither owns the rails. Both own the decisioning. Both are growing faster than the underlying market.
Banks have largely treated the customer screen as the product and left the decisioning fragmented across rail teams. That is not an aesthetic problem. To me, its really the reason the economics are leaking.
The work is harder than the headlines suggest
The work the control plane has to do is getting harder at exactly the same time the rails are getting cheaper. The rails are multiplying. A payment in 2026 might move across SWIFT, FedNow, the issuing bank’s tokenized deposit rail, Broadridge’s distributed ledger repo platform, Partior, Kinexys, a stablecoin on Base or Arc, or a hybrid path that uses several in sequence. Broadridge’s DLR alone processed $384 billion in average daily repo volume in December 2025, totaling close to $9 trillion for the month, and ran 508% year on year growth into January 2026. That is one rail, in one segment.
Now we have the instruments multiplying alongside the rails. Tokenized deposits at JPMorgan, HSBC, Citi, DBS, and various regional consortia. Stablecoins at $315 billion in circulating supply at the end of Q1 2026, with USDC at roughly $78 billion and the GENIUS Act rulemaking in train. Tokenized money market funds. Tokenized treasuries. Each instrument has its own settlement finality, its own legal framework, its own regulatory regime, its own operational hours, its own credit characteristics. The number of permutations a single payment might take is exploding, and each permutation puts a different combination of counterparty and compliance risk on the table.
The orchestration layer is what keeps that complexity legible to the client. Without one, the client makes rail and instrument choices on every transaction, which is exactly what the treasurer’s job is not. With one, the rails and instruments become fungible from the client’s point of view, and the bank’s value capture moves from operating any particular rail to operating the layer that routes across all of them.
Where the AWS analogy breaks
The cloud parallel is worth pushing on, because the economics it predicts are not subtle. In 2010, servers and storage were the expensive part of the IT stack. Enterprises bought hardware, depreciated it, and staffed teams to run it. The control layer was thin and often built in house. By 2025, the economics had inverted. Physical infrastructure became a commodity. The control plane became the product. The hyperscalers captured the value migration not because their servers are better, but because the coordination layer they operate is hard to replicate and gets stronger with each additional service they integrate.
The directional analogy holds for payments. The specifics break in three places that matter for any bank trying to position itself.
The first break is balance sheet. AWS commoditized compute because compute is fungible. A virtual CPU running a workload in Oregon is interchangeable with one running it in Ohio. Of course, money is fungible, but not quite fungible in that way. A dollar in a tokenized JPMorgan deposit account carries JPM’s credit. A dollar in USDC carries Circle’s credit and the credit of its reserve managers. A dollar in a stablecoin issued by a smaller player carries different counterparty risk again. The control plane in payments cannot route across these instruments without taking a view on the credit of each one, and the credit of each one is bound to a balance sheet the orchestration layer doesn’t own. Pure play orchestration fintechs hit this ceiling fast. Airwallex and Wise can route around correspondent banking elegantly, but they cannot route a $500 million corporate intragroup transfer the way Kinexys can, because Kinexys is implicitly backed by JPM’s balance sheet. When Mitsubishi adopted Kinexys for intragroup cash management in March 2026, the single transaction limit was $500 million. No fintech in the world can write that ticket! The control plane does not exist as a separate economic layer in that transaction. It is fused to the underwriter, and the underwriter is the bank.
The second break is the difference between routing and clearing. The orchestration layer chooses the rail. It does not, by itself, eliminate settlement risk. A control plane that picks a path and pushes a transaction through is doing routing. Clearing is what happens when obligations between counterparties are netted, validated, and finalized, and historically that has required either a balance sheet between them (correspondent banking) or a multilateral utility above them. CLS Bank settles over $8 trillion a day in FX, with peak days above $19 trillion, and reduces funding requirements by more than 96% through multilateral netting. That capital efficiency is a clearing achievement, not a routing one. It exists because CLS nets multilateral obligations across a closed set of currencies and members, with explicit settlement rules and loss-mutualization. The original lesson from Herstatt in 1974 was that settlement risk is what kills you when you assume the routing was the hard part.
Atomic settlement networks like Partior collapse routing and clearing into a single primitive, which is why their growth is real and why the founding banks invested. Atomic PvP eliminates the temporal gap that creates Herstatt risk in the first place. A routing engine sitting above multiple non atomic rails does not. It sequences transactions across systems that each have their own settlement finality, their own legal framework, and their own failure modes. The economics of orchestration are real. They are not the same as the economics of clearing. A bank that builds an elegant control plane on top of fragmented rails has built something useful, but it has not built CLS, and treating the two as equivalent is the mistake that gets noticed when something breaks.
The third break is concentration risk. The same property that makes a control plane valuable, coordination of complex systems through a single decisioning point, also makes it dangerous when it fails. The October 2025 AWS outage took down a long list of banks and fintech services that had not realized how dependent they had become on a single coordination layer. The FFIEC has been writing about cloud lock in risk since 2020 and the BIS has flagged it repeatedly. The same risk pattern transfers directly to payments orchestration. If a bank’s flow runs through a single proprietary control plane, an outage in that plane is an outage in the bank’s payments business. If an industry’s flow concentrates into a small number of orchestration providers, a failure in one becomes systemic. This is not theoretical. It is why regulators are watching the orchestration question, and why any bank thinking about ceding the layer to a third party should think hard about what operational resilience looks like in that world.
Who is trying to own the layer
The right way to read the current wave of institutional announcements is not which bank is issuing which instrument, but which institution is making a credible run at the decisioning layer above the instruments. Four categories of competitor are visible. A fifth is harder to name but is already taking flow.
The first is the incumbent network operators, with SWIFT the cleanest example. SWIFT completed the design phase of its blockchain shared ledger in March 2026 and is targeting a live MVP running tokenized deposit payments before the end of the year, with more than 40 banks participating. The architecture is explicitly orchestration. SWIFT will operate the ledger, validate funding commitments, coordinate interbank workflows, and reuse existing compliance processes. The phrase “shared digital orchestration layer” appears in the official announcement. SWIFT is making a serious bid to migrate from messaging layer of the old rails to coordination layer of the new ones, with the same network of 200-plus jurisdictions. If that succeeds, SWIFT preserves its position. If it fails, SWIFT faces the harder problem that messaging and decisioning are different businesses with different economics.
The second is the mega bank platforms. Kinexys is the most visible, processing roughly $7 billion a day with cumulative volume above $3 trillion since launch in 2020, and a public target of $10 billion daily. JPMorgan, Citi, HSBC, BNP Paribas, and a handful of others are building similar capability. The advantage is structural. The bank’s balance sheet is already underneath the platform, which means liquidity and FX are native rather than bolted on. Demand from the bank’s existing corporate base is built in. The disadvantage is that a single-institution platform has interoperability limits unless it opens up. Kinexys has started doing exactly that, with cross-chain settlement work and a corporate pipeline that now extends to Mitsubishi, Siemens, Brevan Howard, BlackRock, and LSEG. Whether mega banks can credibly run an industry-wide orchestration layer, or whether they end up running multiple competing single-institution layers, has not been settled. The answer matters for everyone else in the stack.
The third is the consortia. Partior, founded by JPMorgan, DBS, and Standard Chartered, with Deutsche Bank and Emirates NBD now investors, runs 24/7 atomic multi-currency settlement and describes itself as an “interoperable neutral network” enabling orchestration across platforms. Deutsche Bank’s head of cash management, after the bank’s first euro-denominated transaction on Partior in late 2025, framed the entire strategy in orchestration terms
“a future using multiple rails, be it SWIFT, Stablecoins, or blockchain based solutions, where intelligent and negotiated routing produces maximum value.”
Project Agorá at the BIS Innovation Hub is doing similar work for the official sector. Consortia have legitimate neutrality, which is their main advantage, and slow governance, which is their main constraint. They tend to win where no single bank has pricing power and lose where one does.
The fourth is the fintechs. Airwallex, Wise, Stripe, BVNK, and a long list of specialists are building orchestration top down, starting from client experience and acquiring rail access as they grow. The advantage is design discipline and speed. The disadvantage, as covered above, is that they hit a balance sheet ceiling at a certain transaction size and a regulatory ceiling at a certain flow type. This category is taking the most market share at the SMB and mid market levels, and is the most likely to be acquired or partnered by a bank, consortium, or payment network that wants the technology and does not have time to build it.
A fifth category, harder to name, is the stablecoin issuers. When a stablecoin becomes the settlement asset for a payment, the issuer is part of the orchestration fabric whether it markets itself that way or not. Circle’s CCTP is explicitly an orchestration network for USDC across chains. Tether’s dominance in certain emerging market corridors is a de-facto orchestration layer that bypasses banks entirely. Visa is now settling stablecoin volume at a $4.5 billion annualized run rate. The GENIUS Act rulemaking over the next few months will set the terms on which this category competes against the bank led options.
Infrastructure inversion, again
For two decades the valuable part of financial services sat at the application layer, the consumer fintech UX, the small business front door, the embedded finance integration. Most venture investment in the sector flowed there because that was where the customer facing differentiation lived. The migration has reversed. Value is now moving back down to the infrastructure layer, and I think specifically to the orchestration layer that sits just above the rails and instruments. The companies that own that layer will be the picks and shovels of the next decade of financial services, the way the hyperscalers became the picks and shovels of the software industry.
The orchestration layer in payments is not perfectly analogous to the cloud control plane, because money is not perfectly analogous to data. Balance sheet, clearing finality, and concentration risk shape who can credibly own the layer and how much of it any one institution should try to own. A bank that participates only as a rail operator, in a world where the orchestration layer is forming above it, ends up in the position of a hardware company in a software industry. The rail will be used, but the revenue will be thin and the client relationship will sit one layer up.
The right question for a bank strategy review in 2026 is really not whether to do onchain payments, or even whether to issue a tokenized deposit. It is what position the bank will hold in the orchestration layer that is forming. Operate one or more rails and cede the layer. Participate in a consortium and share it. Or build a proprietary version that can eventually open up. None of those is to sit out. The layer is being built and the question is which seat at which table.
Clients buy certainty. Rails deliver commodities. The orchestration layer is where certainty gets constructed, and the institutions that figure out how to construct it without becoming a single point of failure are the ones that will get paid.
References
AWS Q4 2025 earnings, CNBC: $35.6B revenue, 24% YoY growth, $142B run rate, 35% margin
Swift’s blockchain-based shared ledger progresses to MVP, SWIFT: “shared digital orchestration layer,” March 2026
Kinexys by J.P. Morgan: $7B daily, $3T cumulative since 2020, $10B daily target
Mitsubishi Corporation adopts Kinexys Digital Payments, JPMorgan: March 2026
Broadridge DLR processes ~$9T in December, Broadridge: $384B average daily volume
Broadridge DLR achieves 508% YoY growth in January, Broadridge
Deutsche Bank conducts first euro transaction via Partior, Deutsche Bank: September 2025
Partior: JPMorgan, DBS, Standard Chartered, Deutsche Bank, Emirates NBD
Airwallex 2025 end-of-year mission update: $266B annualized volume, end-2025
Airwallex revenue and valuation, Sacra: $1.2B ARR, March 2026
Wise plc investor relations: £181.7B FY26 cross-border volume, 75% of transactions instant, Wise Platform ~5% of volume
Total stablecoin supply, DefiLlama: $315B end Q1 2026
Visa stablecoin settlement statistics, Stablecoin Insider: $4.5B annualized run rate, January 2026
CLSSettlement, CLS Group: ~$8T daily settlement, 96% funding reduction via multilateral netting
The day the internet broke banking, The Financial Brand: October 2025 AWS outage analysis