BANKING 2035 | PART 1 OF 2 Who Holds the Money, Who Holds the Customer

 

AI, the Finternet and stablecoins: four scenarios for banks, and what their boards must decide now

Five propositions to argue with

This paper is written to provoke debate in boardrooms, not to settle it. Each proposition below is argued in the sections that follow. Readers who disagree are invited to say why.

•      Your AI programme will not make your bank more profitable. It will only keep it in the game. Most of the savings from AI are likely to be competed away to customers (Section 1).

•      India’s deposit franchise is already eroding, and crypto has nothing to do with it. The share of low-cost current and savings deposits has fallen from 44% to about 39% (Section 3).

•      The most likely future for Indian banks is to become excellent suppliers to interfaces they do not own. UPI has already shown how (Section 4).

•      The next bank run may be started by software, not by people. AI agents can move money faster and more in step than any queue of depositors (Section 2).

•      A bank that cannot prove it is secure and trustworthy will soon not be allowed to connect, or chosen. In an open, agent-driven system, resilience and trust become conditions of growth (Section 7, and Part 2).

Executive summary

By 2035, a bank’s value will depend less on how well it uses AI than on two things it does not fully control: whether money stays on bank balance sheets, and whether customers still make their financial decisions inside the bank’s own channels. AI, tokenisation and stablecoins are pushing both questions open at the same time.

Three forces, two uncertainties. Agentic AI could cut bank operating costs by 20% or more, but most of that saving is likely to be competed away to customers. The Finternet, the vision of interconnected tokenised ledgers set out by Agustín Carstens and Nandan Nilekani, is being built at wholesale level: the BIS has shown that tokenised deposits and central bank reserves can settle cross-border chains atomically. Stablecoins, legal in the United States since July 2025, compete for the deposit itself. Together these forces leave two genuine uncertainties: who holds the money layer, and who holds the customer interface.

Four scenarios for 2035. Crossing these gives four worlds. In Fortified Incumbents, banks absorb the new technology. In Public Rails, Private Balance Sheets, open networks and AI agents own the customer while banks supply regulated balance sheets. In Banks as Token Houses, tokens spread but banks issue and custody them. In Wallet World, big-tech and stablecoin wallets hold balances and banks shrink to narrow roles.

India is not starting where the West is. UPI has already separated the payments interface from the bank. The Account Aggregator network now facilitates 8.4% of retail and MSME lending by value. The RBI has rejected stablecoins in favour of CBDCs and is piloting tokenised deposits on its wholesale CBDC. Indian banks are also losing low-cost deposits for reasons unrelated to crypto: the CASA share of deposits has fallen from 44% to about 39% as households move savings into markets. On current policy India is heading for the second scenario. It keeps the banking system intact but compresses margins in the bank’s most profitable lines.

The US picture shifted this autumn. The OCC’s final GENIUS Act rule (October 2026) tightened the ban on stablecoin yield, the CLARITY Act failed in the Senate in September, and the November midterms are reported to have given Democrats control of the Senate. Near-term pressure toward Wallet World in the US may have eased, while bank-led tokenisation, as in Hong Kong where HSBC and a Standard Chartered joint venture received the first stablecoin licences, has gained ground.

Three foundations decide who wins. Whichever world arrives, defence, trust and data will decide whether Indian finance captures the gains or absorbs the losses. AI-enabled attacks are already draining smaller lenders, trust is still treated as a given, and India’s richest asset, the data flowing through its public rails and sector stacks, is not yet usable for AI. Section 7 summarises these issues; Part 2 of this paper, Building the Foundations, examines them in depth and sets out what regulators and the ecosystem should build.

What boards should do. Seven moves pay off in every scenario: take AI cost-out to the core of the bank; build tokenised-deposit readiness; make the bank usable by AI agents; take part in open credit networks now, at controlled scale; re-price the deposit franchise and liquidity buffers on the assumption that customer inertia disappears; bring AI, cyber and quantum resilience onto the board’s own agenda; and make trust visible and verifiable. Larger commitments should be held as options, triggered by the signposts in Section 6 and by the bank’s own results. Section 9 poses the questions boards should be debating now.

1. Why the usual framing misleads boards

Most board discussion of the future of banking treats AI as the main event and asks how fast the bank can adopt it. The question matters, but it is the wrong place to start.

AI is a cost curve, not a strategy.  When every bank can buy the same models from the same vendors, AI-driven efficiency becomes table stakes. McKinsey estimates agentic AI could cut operating costs by 20% or more, yet expects most of that gain to be competed away. It projects that global bank profit pools of about $1.2 trillion could shrink by up to 10% over five to ten years if business models do not change. The bank that fails to cut costs loses; the bank that succeeds mostly keeps pace.

The real exposure is the franchise. Banks earn their best returns from two quiet advantages: customers who leave money in low-yield accounts out of inertia, and customers who begin every financial decision in the bank’s own channel. Both advantages are under pressure from forces outside the bank. Customer-side AI agents can sweep idle balances to the best rate every night. Open networks such as UPI already route payments through third-party apps. Stablecoins and tokenised money-market funds offer a dollar or yield-bearing balance outside the deposit system.

Scenarios, not forecasts. No one can tell a board which of these pressures will dominate by 2035, because the outcome turns on regulatory choices still being made in Washington, Mumbai, Hong Kong and Brussels. This paper therefore builds scenarios around the two variables that most change a bank’s income statement, and asks what a board should do now that holds up in all of them.

2. Three forces reshaping banking

2.1 AI and the arrival of agents

Executive expectations are running ahead of results. In Accenture’s 2026 survey, 57% of banking executives expect AI agents to be fully embedded in risk, compliance, audit and fraud within three years, and 56% expect broad adoption in credit, lending and KYC. BCG estimates agents could raise bank profitability by 30% and cut costs by 30 to 40% by 2030. Yet McKinsey reports that about 80% of companies using generative AI see no significant bottom-line impact so far. All of these are projections or expectations, not measured outcomes.

The more consequential shift is on the customer side. Once a customer’s AI agent can compare products, move money and apply for credit on the customer’s behalf, the bank loses both of its inertia advantages. India is already building the rails. In 2025, NPCI, Razorpay and OpenAI piloted payments from inside ChatGPT using UPI Circle and Reserve Pay. In September 2026, Reuters reported that NPCI was weighing a Unified Agent Protocol that would let agents make small payments without approving each one, with liability following whichever party controls the step that failed.

Agents also change the speed of risk. In March 2023, customers withdrew $42 billion from Silicon Valley Bank in a single day and queued a further $100 billion for the next, together about 81% of its deposits, according to Federal Reserve Vice Chair Michael Barr. That run was driven by people using phones. AI agents acting on rate or risk signals could move balances faster and more in step, so liquidity buffers and stress tests calibrated on historical outflow rates may understate the risk. AI also makes fraud harder to stop: synthetic identities and voice or video impersonation weaken authentication at the very moment banks must start authenticating machines as well as people.

2.2 Tokenisation and the Finternet

Carstens and Nilekani’s 2024 BIS paper describes a Finternet of “multiple financial ecosystems interconnected with each other, much like the internet”, built on tokenisation and unified ledgers. What was a vision in 2024 is now a working wholesale prototype. Project Agorá, run by the BIS and the Institute of International Finance with seven central banks and more than 40 private institutions, began user testing in January 2026. In May 2026 it reported that tokenised central bank reserves and commercial bank deposits can settle multi-currency cross-border chains atomically, that settlement finality is legally achievable in all seven jurisdictions, and that tokenisation does not change the legal status of deposits. The next phase is real-value testing.

This is a Finternet that banks inhabit rather than one that replaces them. The pressure falls on revenue lines that exist because ledgers do not talk to each other: correspondent banking, FX spreads, custody, trade documentation and reconciliation-heavy operations. The opportunity lies in programmability: conditional payments, collateral that moves in seconds, and fractional ownership of assets that are illiquid today. The largest US banks are planning a tokenised deposit network through The Clearing House for 2027, explicitly positioned against stablecoins. India’s RBI launched a deposit tokenisation pilot on its wholesale CBDC in October 2025 and a tokenised certificate of deposit pilot in 2026.

For India, the larger long-run opportunity may lie in the retail side of the Finternet that Carstens and Nilekani describe. Tokenised property titles, invoices and receivables could give MSMEs and households collateral that lenders find hard to use today, which would widen the credit market rather than just re-plumb it.

2.3 Stablecoins and the fight for the deposit

The US GENIUS Act, signed on 18 July 2025, legalised payment stablecoins but bars issuers from paying interest or yield. The stablecoin market stood at about $308 billion in August 2026, below its May peak of $322 billion, with Tether and Circle holding about 82% between them. The BIS 2025 Annual Economic Report argued that stablecoins fail three tests of sound money: singleness (they do not always trade at par), elasticity (they cannot expand to meet liquidity needs) and integrity (they are exposed to illicit use).

For banks, the yield ban is the dam. A stablecoin that pays no interest is a payments instrument. One that pays yield, directly or through an affiliate or platform, is a deposit substitute without deposit insurance or reserve requirements. Standard Chartered estimates that if stablecoins reach $2 trillion by end-2028, about $500 billion could leave developed-market bank deposits and about $1 trillion could leave emerging-market banks. Today’s market is about 15% of that threshold, so these are scenario numbers, not forecasts.

Jurisdiction	Position as of November 2026	Implication for banks
United States	GENIUS Act in force; OCC final rule issued 21 October 2026, effective 18 January 2027, bars yield paid in connection with holding a stablecoin and drops the proposed safe-harbour presumption. CLARITY Act failed cloture 49 to 50 on 15 September 2026.	Yield-bearing stablecoins held back for now; enforcement and future rulemaking decide the edges
Hong Kong	First stablecoin issuer licences granted 10 April 2026 to HSBC and Anchorpoint (a Standard Chartered, HKT and Animoca Brands joint venture)	Banks are the first licensed issuers: an early sign of Scenario C
India	RBI warns of risks to monetary sovereignty and prefers CBDC; Finance Minister said in October 2025 that stablecoins are transforming money and capital flows	Domestic stablecoins unlikely near term; policy debate between government and RBI remains open

Table 1. Stablecoin regulation in three reference jurisdictions. Sources: Hogan Lovells, FinTech Weekly, Charltons, ThePrint, Yahoo Finance (see Sources).

3. India’s starting point

India enters this transition with two features no other large economy shares: public digital rails that already sit between banks and their customers, and a central bank that has chosen CBDC over stablecoins. It also enters from a position of balance-sheet strength. Scheduled commercial banks reported a return on assets of 1.4%, return on equity of 13.5% and gross NPAs of 2.2% for 2024-25, the last a multi-decade low, according to the RBI’s Trend and Progress report.

Indicator	Latest figure	What it signals
UPI	25.47 billion transactions worth Rs 31.53 lakh crore in October 2026; volume up 23% year on year	The payments interface already sits outside banks
Account Aggregator	Rs 3.82 lakh crore across 3.68 crore loans in FY26; 8.4% of retail and MSME lending by value	Open data rails reach credit at scale
Unified Lending Interface	64 lenders and more than 136 data services by December 2025	Credit origination is being opened next
GenAI in Indian lenders	86% have use cases under implementation in 2026, up from 10% in 2024	Adoption is fast; scaling to profit is the gap
CASA share of deposits	About 39%, down from a peak of 44% in FY22	The low-cost deposit franchise is already eroding
Credit-deposit ratio	82.5% in June 2026, up from about 75% in mid-2025	Banks are paying more for funding
Retail e-rupee	Rs 772 crore in circulation, March 2026, down 24% year on year	Retail CBDC has not found a use beyond UPI
Tokenised money	Deposit tokenisation pilot on wholesale CBDC since October 2025; tokenised CD pilot on the Unified Markets Interface	Bank money is being tokenised under RBI control

Table 2. India’s digital and banking baseline. Sources: Free Press Journal, Sahamati via IANS, MediaNama, KNN India, Indian Express via SuperKalam, RBI via inkl and NewKerala, Business Standard.

The deposit squeeze is already here. Indian banks are losing low-cost deposits without any help from stablecoins. Households are moving savings into equities and mutual funds, and banks are filling the gap with term deposits and certificates of deposit that cost roughly 7 to 8%, against 3 to 4% for CASA. This matters for the scenarios: Indian customers have already shown they will move money for return once switching is easy. AI agents will make it easier still.

AI adoption is broad but shallow. The BCG, FICCI and IBA study Winning in the AI Era finds generative AI under implementation at 86% of Indian lenders in 2026. The global gap between pilots and profit applies here too; there is little public evidence yet of AI moving Indian banks’ cost-to-income ratios. The RBI’s FREE-AI framework (August 2025) sets out seven principles, including trust, fairness, accountability and explainability, and 26 recommendations. It signals that the RBI will enable AI but hold regulated entities, and therefore their boards, accountable for its use.

Payments taught banks a lesson. UPI let banks keep the deposit while third-party apps took the customer interface, and zero merchant discount rates removed most of the payments revenue. The same pattern could repeat in credit and savings as ULI, the Account Aggregator network and ONDC’s financial services layer mature.

Policy direction is clear, with one open seam. The RBI’s December 2025 Financial Stability Report warned that stablecoins pose significant risks to monetary sovereignty and urged countries to prioritise CBDCs. The RBI is pursuing wholesale and programmable uses instead: tokenised deposits and certificates of deposit, and programmable food subsidies in three states. Because the government has signalled more openness than the RBI, a regulated rupee token, issued or backed by banks, remains a possible compromise by the end of the decade.

4. Four scenarios for 2035

The scenarios cross two uncertainties. The money layer asks whether transactional balances stay mostly in regulated bank money (deposits, tokenised deposits, CBDC) or fragment into stablecoins and other non-bank tokens. The customer interface asks whether customers still decide inside bank channels or delegate to third parties: AI agents, or apps running on open networks such as UPI, ULI and ONDC. AI and tokenised settlement are present in all four worlds. What changes is who captures their value.

Figure 1. Scenario frame for 2035: money layer by customer interface.

A. Fortified Incumbents

Regulators hold the line on stablecoin yield and keep agent access to accounts tightly controlled. Banks deploy AI across operations and launch tokenised deposits through consortia. The franchise survives, but competition passes most AI savings on to customers, so returns improve only modestly. Large banks with scale in technology win; mid-sized banks consolidate.

B. Public Rails, Private Balance Sheets

The UPI pattern spreads from payments to credit, savings and investments. AI agents, running over open protocols such as NPCI’s proposed agent layer, ULI and ONDC, choose products for the customer in real time. Banks remain the regulated holders of deposits and credit risk but compete as suppliers behind interfaces they do not own. Volumes grow sharply, especially in MSME and first-time credit, while margins on deposits and fee products compress. Banks that are the fastest, cheapest and most reliable to call through an API take share.

C. Banks as Token Houses

Stablecoins and tokenised assets spread, but banks keep the customer by becoming the trusted issuers, custodians and distributors. Banks hold stablecoin reserves, issue their own tokenised deposits or licensed stablecoins, and sell tokenised funds and real assets through their own channels. Net interest income falls as balances move into tokens; fee and custody income partly replace it. First movers in digital-asset custody and tokenisation win. Hong Kong’s first licences, both to bank-led issuers, show this path is real.

D. Wallet World

Stablecoins win a yield-like return through platform rewards or new legislation, and big-tech or stablecoin wallets with built-in AI agents become the primary financial account. Core deposits leak, banks shrink toward narrow credit, custody and reserve-holding roles, and much credit moves to funds and private credit. For emerging markets this world also brings dollarisation through the phone. In India, the more plausible route to D may not be dollar stablecoins but large payment apps, which already hold the UPI interface, adding deposit-like products and credit at scale.

Where major systems sit today

India may be heading for B: its public rails are already taking the interface, while the RBI keeps money inside the regulated system. The United States has moved away from D since September 2026, with the CLARITY Act stalled and the OCC tightening the yield ban, and is now drifting between A and C as large banks build tokenised deposit networks. Hong Kong is the clearest early case of C. The euro area looks closest to A, given its emphasis on a digital euro and bank-centred payments, though this paper has not examined it in depth.

5. What each scenario does to the income statement

Table 3 shows the direction of change by 2035 relative to today. All cells are reasoned judgements, not modelled estimates. Quantifying them for a specific bank is the natural next step for any board using this paper.

Line item	A. Fortified Incumbents	B. Public Rails	C. Token Houses	D. Wallet World
Net interest margin	Slightly lower	Sharply lower on deposits	Lower	Sharply lower
Deposit base	Stable	Stable in total, more rate-sensitive	Partly moves into tokens	Shrinks
Loan volumes	Steady growth	Strong growth through open origination	Steady	Shifts to funds and private credit
Fee income	Stable	Lower per unit, higher volume	Higher: custody, issuance, distribution	Lower
Operating cost	Much lower (AI)	Much lower (AI, shared rails)	Lower, with new technology spend	Lower, on a smaller base
Cross-border and FX revenue	Lower	Lower	Lower	Much lower
Return on equity	Modestly higher for leaders	Polarised: winners up, laggards down	Higher for first movers	Lower for most

Table 3. Direction of change in key income statement lines by scenario, 2035 versus today.

Two patterns run across all four columns. First, cross-border, FX and reconciliation-based revenue falls everywhere, because tokenised settlement removes the frictions these lines are paid for. Second, AI cost savings appear everywhere but are kept only where the bank still controls the customer. In scenarios B and D, the savings largely flow to customers and platforms.

6. Signposts to watch, 2026 to 2030

Scenarios become useful when a board knows which signals would show one of them taking hold. Each signpost in Table 4 is observable and tied to a direction on one axis.

Signpost	Points toward	Status, November 2026
US legislation or rules permit yield-like stablecoin rewards	Money fragments (C, D)	Moving the other way: OCC final rule tightens yield ban; CLARITY Act failed in Senate
Stablecoin market passes $1 trillion	Money fragments (C, D)	About $308 billion, August 2026
Banks become licensed stablecoin issuers in major centres	Banks as Token Houses (C)	Hong Kong: HSBC and a Standard Chartered joint venture licensed, April 2026
US bank tokenised deposit network goes live	Bank money holds (A, C)	Planned for 2027
Agorá moves to real-value cross-border settlement	Tokenised bank money (A, C)	User testing from January 2026; real-value phase announced May 2026
NPCI formalises an AI agent protocol for UPI	Agents own the interface (B)	Under consideration, per Reuters, September 2026
Account Aggregator share of retail and MSME lending passes 20%	Open origination (B)	8.4% by value in FY26
RBI extends tokenised deposits or a rupee token to retail use	Tokenised bank money in India (A, C)	Wholesale pilots only
Big tech gains a banking or credit licence in India	Agents and wallets (B, D)	No public indication
Indian CASA share falls below 35%	Deposit inertia breaking (B, D)	About 39%, mid-2026

Table 4. Signposts for scenario tracking.

Indian boards should review the India signposts quarterly and treat the US and Hong Kong ones as early warnings. Dollar stablecoins can reach Indian savers and exporters through offshore channels whatever the RBI decides, so US rules on yield shape India’s exposure too.

7. Beyond the scenarios: the foundations question

The scenarios ask who will capture value. A prior question is whether the system that value flows through can be defended, trusted and powered by the right data. Part 2 of this paper examines these issues in depth; the essentials for boards are below.

•      Defence. AI-assisted attacks reportedly took nearly Rs 300 crore from smaller NBFCs in about two months in 2026. Quantum computing threatens the encryption the whole system relies on, and the G7’s financial cyber experts expect the most critical systems to move to quantum-safe cryptography around 2030 to 2032. In an interconnected system, a partner’s breach becomes your problem (Part 2, Section 2).

•      Trust. In a 2026 EY survey, 82% of Indian banking customers trusted their bank with their data, but only 73% believed fraud is resolved effectively. Once AI agents choose providers, trust that is visible and machine-readable can become a funding advantage (Part 2, Section 3).

•      Data and models. India is far behind in frontier AI: its public AI mission has about $1.2 billion over five years, against roughly $725 billion of capital spending planned by four US technology companies in 2026 alone. Its advantage lies in data from public rails and sector stacks, which a federated institution could make usable for credit, fraud and risk models (Part 2, Sections 4 and 5).

•      The Global South. Components of India’s digital public infrastructure are being piloted in 25 countries. Banks that learn to build on open rails at home can carry those capabilities abroad (Part 2, Section 6).

What it means for boards. Two no-regret moves in Section 8, on resilience and on trust, follow directly from these issues. Several options, including contributing to a shared data institution and lending on sector stacks, depend on whether the ecosystem builds the foundations that Part 2 proposes. Boards have a stake in that outcome and a voice in shaping it.

8. The board agenda

Boards cannot choose their scenario, but they can avoid being caught unprepared by any of them. The agenda below separates commitments that pay off in all four worlds from options to hold until a signpost fires.

No-regret moves

1.    Take AI cost-out to the core, now. Credit operations, KYC, compliance, collections and servicing. In every scenario the savings are eventually competed away, and the bank that has not made them loses outright. Measure progress in cost-to-income terms, not in the number of pilots.

2.    Build tokenised-deposit and programmable-payment readiness. In India, this means active participation in the RBI’s deposit tokenisation and wholesale CBDC pilots, and core banking that can hold and move tokenised balances. The cost of readiness is small next to the cost of arriving late, which risks losing custody and settlement roles.

3.    Make the bank callable by agents. Machine-readable products and prices, reliable APIs, agent authentication and clear liability terms. In scenarios B and D an AI agent chooses the bank, and it will choose one it can read and call reliably.

4.    Take part in open credit networks now, at controlled scale. Connect to the Account Aggregator network, ULI and ONDC’s credit layer, with explicit limits on exposure, pricing and risk. Treat it as a way to learn the unit economics and build real-time underwriting before competitors lock in the borrower data. Banks already account for 47.3% of AA-enabled lending by value in the second half of FY26, so this is no longer an early bet. The risk to manage is adverse selection: when a loan request reaches many lenders at once, a bank’s wins can skew toward borrowers others priced higher or declined.

5.    Re-price the deposit franchise and liquidity on behaviour, not inertia. Stress-test net interest margin against a world where customers’ agents sweep idle balances daily, and recalibrate liquidity buffers for outflows at agent speed. India’s falling CASA share suggests this is a present risk, not a 2035 one.

6.    Put AI, cyber and quantum resilience on the board’s own agenda. The FREE-AI framework and the RBI’s 2026 cybersecurity directions place accountability on regulated entities. Boards need a named owner for AI model risk, an inventory of AI uses, a view on liability for agent-initiated transactions, and a funded plan to build a cryptographic inventory and move critical systems to quantum-safe encryption. Smaller institutions should actively seek shared services rather than wait for them. Part 2, Section 2 sets out the shared capabilities they will need.

7.    Make trust visible and verifiable. Adopt verified identity across domains, apps and agents; publish service and fraud-resolution performance; and make security and consent practices machine-readable so that customers’ AI agents can recognise them. Treat this as a source of funding advantage, not a compliance cost. Part 2, Section 3 describes what a financial trust standard could contain.

Options to hold

•      Balance-sheet-as-a-service as the core model. Reorganising the bank around supplying credit and deposits behind open networks and fintech front ends, with volume rather than margin targets and less own distribution. This cannibalises own channels if the bank turns out to be in Scenario A or C. Trigger: the bank’s own open-network loans perform at least as well as its own-channel book over a full credit cycle and add to return on equity.

•      Token issuance and custody. Issuing tokenised deposits or a regulated rupee token, and offering custody for tokenised assets or stablecoin reserves. Trigger: the RBI opens a retail tokenised-money path, or Agorá-style settlement goes live in rupee corridors.

•      A proprietary customer agent. Building the bank’s own AI agent is worth it only if banks keep the interface. Trigger: regulators restrict third-party agent access to accounts.

•      Contributing to a shared financial AI data institution. Making de-identified data available to secure, federated training environments in return for access to better credit and fraud models. Trigger: a governance, security and liability framework that the bank’s board is satisfied with. See Part 2, Section 4.

•      Lending and insurance built on sector stacks. Agricultural, health and green-energy products underwritten with Agri Stack, ABDM and India Energy Stack data through ULI and the Account Aggregator network. Trigger: consent and data-quality standards that make this data reliable enough for underwriting. See Part 2, Section 5.

Questions every board should ask management

•      What share of our net interest margin depends on customer inertia, and what happens to it if that inertia is gone by 2030?

•      Which of our revenue lines exist only because ledgers do not talk to each other?

•      If an AI agent were choosing a bank for our best customer today, would it choose us, and why?

•      How fast could our deposits leave if customers’ agents moved them, and do our liquidity buffers assume that speed?

•      If a smaller lender or fintech partner connected to us were breached tomorrow, how quickly would we know, and what would it cost us?

•      Which of our systems would be exposed if today’s encryption were broken, and when will they be migrated?

•      If our main AI model provider changed its pricing or access tomorrow, what would stop working, and how quickly could we switch?

•      Who on this board can challenge management on AI model risk and tokenisation architecture?

9. Questions for discussion

These questions do not have settled answers. They are offered for boards, management teams and industry forums to argue over.

•      If customers’ AI agents could move every rupee to the best rate overnight, what would your deposit base look like a year later?

•      Would your bank rather own fewer customers at a healthy margin, or serve everyone’s interface at a thin one? Has the board actually made that choice, or is it being made by default?

•      Which of the four scenarios is your current strategy implicitly betting on, and what would make you change that bet?

•      UPI made payments free and moved the customer to third-party apps. Should Indian banks welcome the same model in credit and savings, or resist it?

•      Should India allow banks to issue rupee tokens to retail customers before someone else offers Indian savers a dollar token on their phones?

•      When a customer’s AI agent makes a bad financial decision, who should bear the loss: the customer, the agent’s provider, the bank or the network?

•      Is your bank spending on AI to cut costs, to win customers or to defend itself? Can it afford to do only one?

10. Limitations and next steps

•      Most AI impact figures are consultancy projections and executive expectations. Indian bank-level evidence of AI-driven cost reduction is not yet public.

•      The income statement effects in Section 5 are directional. A next version should model net interest margin at risk for an illustrative Indian bank under each scenario.

•      Several figures come from secondary reporting of primary documents (RBI, Sahamati, NPCI, BIS, BCG-FICCI-IBA).

•      US political developments after the November 2026 midterms are reported, not yet reflected in legislation, and may change quickly.

•      The euro area and other emerging-market DPI systems, including Indonesia, are treated briefly. Defence, trust and data are summarised here and examined in Part 2.

“Innovations like stablecoins are transforming the landscape of money and capital flows. Such developments underscore the scale of transformation underway. They also remind us that no nations can insulate itself from systemic change. Whether we welcome these shifts or not , we must prepare to engage with them.” Ms. Nirmala Sitharaman. Finance Minister, India

About the author

T Koshy is the founding Managing Director and CEO of India’s Open Network for Digital Commerce (ONDC), and serves on the Board of Commissioners of the Indonesia Open Network (ION), a Beckn protocol-based digital public infrastructure initiative modelled on ONDC.

He is a Director on the boards of CRIF High Mark, Aditya Birla Housing Finance, Perfios Account Aggregation Services and Last Mile Care.

His career spans NSDL, EY, Projects with the World Bank and the Asian Development Bank, as well as several national digital transformation programmes in India. He is a distinguished alumnus of IIM Bangalore.

Disclosure: the author’s roles at a credit information company, an Account Aggregator, a housing finance company and an open commerce network relate directly to several scenarios discussed in this paper. The views expressed are his own and do not represent those of any organisation with which he is associated.

Sources

1.      Carstens, A. and Nilekani, N., Finternet: the financial system for the future, BIS Working Paper 1178, April 2024. https://www.bis.org/publ/work1178.pdf

2.      BIS, Project Agorá shows how tokenisation can improve wholesale cross-border payments, media release, 27 May 2026. https://www.bis.org/media-releases/20260527-project-agora-shows-how-tokenisation-can-improve-wholesale-cross-border-payments-work-will

3.      PYMNTS, Central banks hit next stage in Agorá cross-border payment project, January 2026. https://www.pymnts.com/?p=3387557

4.      Cointelegraph, BIS Annual Economic Report 2025: stablecoins fail as money, June 2025. https://cointelegraph.com/news/stablecoins-fail-money-bis-report

5.      McKinsey & Company, Agentic AI will shake up banking, shrinking global profit pools, 2025. https://www.mckinsey.com/industries/financial-services/our-insights/banking-matters/agentic-ai-will-shake-up-banking-shrinking-global-profit-pools

6.      BCG, How retail banks can put agentic AI to work, 2026. https://www.bcg.com/publications/2026/how-retail-banks-can-put-agentic-ai-to-work

7.      Banking Dive, Banks aim for agentic AI scale in 2026 (Accenture Top Banking Trends 2026), January 2026. https://www.bankingdive.com/news/banks-agentic-ai-scale-2026-accenture/809585/

8.      KNN India, Indian banks accelerate GenAI adoption (BCG, FICCI and IBA, Winning in the AI Era), August 2026. https://knnindia.co.in/news/newsdetails/sectors/indian-banks-accelerate-genai-adoption-as-implementation-rises-to-86-report

9.      Reserve Bank of India, Report of the FREE-AI Committee, press release, August 2025. https://www.rbi.org.in/Scripts/BS_PressReleaseDisplay.aspx?prid=61018

10.   KPMG India, RBI’s FREE-AI committee report in the financial sector, September 2025. https://kpmg.com/in/en/insights/2025/09/rbis-free-ai-committee-report-in-the-financial-sector.html

11.   Reserve Bank of India, Report on Trend and Progress of Banking in India 2024-25, press release, December 2025. https://rbi.org.in/Scripts/BS_PressReleaseDisplay.aspx?prid=61902

12.   CNBC, SVB customers tried to pull nearly all deposits in two days, Fed’s Barr testifies, March 2023. https://www.cnbc.com/2023/03/28/svb-customers-tried-to-pull-nearly-all-deposits-in-two-days-barr-says.html

13.   Forkast, India eyes AI-driven UPI payments as NPCI weighs agent protocol, September 2026. https://forkast.news/india-eyes-ai-driven-upi-payments-as-npci-weighs-agent-protocol/

14.   Spark, Stablecoin market cap tracker, August 2026. https://www.spark.money/tools/stablecoin-market-cap-tracker

15.   Spark, The CLARITY Act: how Congress is rewriting the rules on stablecoin yield, June 2026. https://www.spark.money/research/clarity-act-stablecoin-yield-rules

16.   Hogan Lovells, OCC finalizes its GENIUS Act rule, Part 1: permitted activities and prohibition on yield, October 2026. https://www.hlc.com/en/publications/occ-finalizes-its-genius-act-rule-part-1-permitted-activities-and-prohibition-on-yield

17.   FinTech Weekly, CLARITY Act cloture vote fails 49 to 50, September 2026. https://www.fintechweekly.com/news/clarity-act-cloture-vote-fails-49-50-september-2026

18.   KuCoin News, Democrats win 2026 midterms, crypto regulation outlook shifts, November 2026. https://www.kucoin.com/news/flash/democrats-win-2026-midterms-crypto-regulation-outlook-shifts

19.   Cointelegraph, Standard Chartered: stablecoins a real threat to US bank deposits, January 2026. https://cointelegraph.com/news/stablecoins-real-threat-us-bank-deposits-says-standard-chartered

20.   KuCoin News citing the Wall Street Journal, JPMorgan, Citi and major US banks plan tokenized deposit network by 2027. https://www.kucoin.com/news/flash/jpmorgan-citi-and-major-u-s-banks-plan-tokenized-deposit-network-by-2027

21.   Charltons, HKMA issues first stablecoin issuer licences, April 2026. https://www.charltonslaw.com/hkma-issues-first-stablecoin-issuer-licences/

22.   ThePrint/PTI, Adoption of stablecoins could pose risk to monetary sovereignty: RBI, December 2025. https://theprint.in/economy/adoption-of-stablecoins-could-pose-risk-to-monetary-sovereignty-rbi/2816056

23.   Business Standard, RBI to launch deposit tokenisation pilot on wholesale CBDC, October 2025. https://www.business-standard.com/economy/news/rbi-deposit-tokenisation-pilot-cbdc-wholesale-digital-tokens-oct8-125100700532_1.html

24.   Yahoo Finance, Indian banks to pilot tokenized deposits as Finance Minister hints at regulation, October 2025. https://finance.yahoo.com/news/indian-banks-pilot-tokenized-deposits-110301545.html

25.   inkl/Economic Times, Retail CBDC in circulation falls by 24% in FY26. https://www.inkl.com/news/retail-central-bank-digital-currency-in-circulation-falls-by-24-in-fy26

26.   NewKerala, RBI expands e-rupee pilots, starts tokenised certificate of deposit trial, May 2026. https://www.newkerala.com/news/a/rbi-expands-e-rupee-pilots-starts-tokenised-credit-deposit-979.htm

27.   Free Press Journal, UPI transactions hit 25.47 billion in October, November 2026. https://www.freepressjournal.in/business/upi-transactions-hit-2547-billion-in-october-value-soars-to-3153-lakh-crore

28.   IANS, India’s Account Aggregator ecosystem drives Rs 3.82 lakh crore loans in FY26 (Sahamati), August 2026. https://ianslive.in/indias-account-aggregator-ecosystem-drives-rs-382-lakh-crore-loans-in-fy26--20260827133606

29.   MediaNama, Unified Lending Interface: 64 lenders, 136 data services, January 2026. https://www.medianama.com/2026/01/223-united-lending-interface-64-lenders-136-data-services/

30.   SuperKalam/Indian Express, Amid slowing CASA growth, banks turn to costly solutions, July 2026. https://superkalam.com/current-affairs/05-07-2026/amid-slowing-casa-growth-banks-turn-to-costly-solutions-pg13-08393e5c-8693-4908-ac01-27fd8944c0cb

31.   The Week/CareEdge, Indian banks to see credit growth in FY27 but deposits set to lag, March 2026. https://www.theweek.in/news/biz-tech/2026/03/27/indian-banks-to-experience-credit-growth-in-fiscal-2026-27-but-deposits-set-to-lag-behind.html

32.   Inventiva, citing The Economic Times, AI-powered cyberattacks siphon nearly Rs 300 crore from small NBFCs, October 2026. https://www.inventiva.co.in/trends/ai-powered-cyberattacks-siphon-%E2%82%B9300-crore-from-small-nbfcs-as-rbi-flags-risks-to-banks/

33.   K S & K, RBI Cybersecurity, Technology Risk, Resilience and Assurance Framework, August 2026. https://ksandk.com/newsletter/rbi-cybersecurity-technology-risk-resilience-framework-2026/

34.   G7 Cyber Expert Group, Quantum Roadmap for the financial sector, January 2026. https://home.treasury.gov/system/files/136/G7-CEG-Quantum-Roadmap.pdf

35.   NewKerala, Customers trust banks with data, but only fully satisfied with fraud resolution: EY report, May 2026. https://www.newkerala.com/news/a/customers-trust-banks-data-only-fully-satisfied-fraud-782.htm

36.   Press Information Bureau, IndiaAI Mission backgrounder, February 2026. https://static.pib.gov.in/WriteReadData/specificdocs/documents/2026/feb/doc2026210784401.pdf

37.   Tom’s Hardware, Big Tech’s AI spending plans reach $725 billion, April 2026. https://tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion

38.   Open Source For You, MeitY and MEA drive global rollout as 25 nations pilot India’s DPI, July 2026. https://www.opensourceforu.com/2026/07/meity-mea-drive-global-rollout-as-25-nations-pilot-indias-dpi/