JPMorgan Chase moved two billion dollars in AI spending from discretionary innovation to core infrastructure — alongside data centers, payment systems, and risk controls. A CEO calling AI important is marketing. Reclassifying the budget line is an accounting commitment that affects capital ratios, auditor scrutiny, and regulatory reporting. It is the hardest corporate signal to reverse.
In January 2026, JPMorgan Chase disclosed that it had reclassified its AI spending from discretionary innovation to core infrastructure. The two billion dollars the bank allocates to artificial intelligence now sits alongside data centers, payment systems, and risk controls in its nineteen point eight billion dollar technology budget. The change was buried in an earnings presentation, not announced with a press release. That is the point.
A CEO telling CNBC that AI is transformative costs nothing. A press release about an AI initiative costs nothing. A pilot program with fifty employees costs almost nothing. But moving two billion dollars from one budget category to another changes what auditors examine, what regulators scrutinize, what capital ratios reflect, and what the board must approve to reverse. It is the corporate equivalent of pouring a foundation — not because it cannot be undone, but because undoing it requires demolishing what was built on top.
Jamie Dimon has compared AI to the printing press, the steam engine, and electricity. Those comparisons are free. The reclassification is not.
What Accounting Reveals
There is a hierarchy of corporate commitment that runs from rhetoric to restructuring. At the bottom: executive statements. CEOs say things. The statements cost nothing and bind no one. Above that: pilot programs. A company allocates a team, runs an experiment, publishes results. The cost is real but contained — a rounding error on the income statement. Above that: dedicated headcount. JPMorgan has two thousand people working on AI. Hiring is harder to reverse than a pilot but easier than what comes next.
At the top of the hierarchy: reclassification. When spending moves from discretionary to core, it changes the institutional physics. Discretionary budgets are the first to be cut in a downturn. Core infrastructure budgets are the last. The distinction is not semantic. It determines what happens when revenue falls, when regulators ask questions, when the board reviews the annual plan. Discretionary innovation is a bet the company is making. Core infrastructure is a bet the company has made.
JPMorgan's chief financial officer told analysts that legacy modernization investment had peaked, freeing capital for reallocation toward AI. That sentence describes a one-way door. The legacy systems are modernized. The capital has moved. The institutional muscle memory now treats AI spending the way it treats the payment rails and the fraud detection systems — as something the bank cannot operate without.
What Two Billion Dollars Built
The reclassification is not aspirational. It reflects infrastructure that already exists.
JPMorgan's internal LLM Suite — a generative AI platform connected to models from both Anthropic and OpenAI, updated every eight weeks — reached two hundred and fifty thousand employees by the end of 2025. That is more than half the bank's global workforce. The platform started in May 2023 with a small team. By August 2024, sixty thousand employees had access. By September, the rollout expanded to a hundred and forty thousand. The trajectory is not a pilot. It is a deployment curve that looks like email adoption in the 1990s — except compressed into two years.
The bank's Contract Intelligence system reviews twelve thousand commercial loan agreements and extracts three hundred and sixty thousand hours of annual legal processing time. Four hundred AI use cases were in production by the end of 2024. The bank doubled that number in 2025. Teresa Heitsenrether, the chief data and analytics officer JPMorgan appointed in June 2023 to lead this work, described the LLM Suite as functioning like a research analyst that can offer information, solutions, and advice on any topic. American Banker named it Innovation of the Year.
The deployment extends beyond the generalist platform. Connect Coach supports private bank advisors. SpectrumGPT handles specialized workflows. Algorithmic trading applications have expanded. Credit risk assessment, fraud detection, and customer service all run AI systems in production — not as experiments alongside the existing process, but as the process itself.
This is what reclassification looks like from the inside. You do not move spending to core infrastructure because you believe AI will be important someday. You move it because removing the AI would break operations that currently depend on it. The reclassification is not a prediction. It is an acknowledgment.
The Peer Signal
JPMorgan is not alone, but it moved first and furthest.
Bank of America allocates approximately four billion dollars of its thirteen billion dollar technology budget to AI — a higher percentage than JPMorgan but without the formal reclassification to core infrastructure. Its Erica chatbot, launched in 2016, was one of the earliest large-scale AI deployments in banking. The bank reports that AI is boosting banker productivity and revenue but has not disclosed the same structural budget shift.
Goldman Sachs is rolling out a generative AI assistant to bankers, traders, and asset managers. The bank projects that global AI capital spending will reach five hundred and twenty-seven billion dollars in 2026 — a forecast it is acting on by deploying copilots across its own divisions. Morgan Stanley deployed an internal GPT-4-powered assistant to its financial advisors in September 2023, drawing on approximately a hundred thousand research reports. The tool is now embedded in the advisory workflow.
Every major bank is spending on AI. What distinguishes JPMorgan is the accounting treatment. The others are running AI as innovation — significant, growing, strategically important, but still discretionary. JPMorgan has declared it infrastructure. The distinction matters because it changes what happens next. Innovation budgets can be redirected. Infrastructure budgets, by institutional design, cannot.
The Historical Pattern
Infrastructure gets reclassified when adoption is genuine. Bubbles live in the innovation line until they pop.
The American railroad system was treated as speculative investment through the 1860s and 1870s. Banks financed railroads as venture bets — high-risk, high-reward propositions with uncertain demand. By the 1880s, railroads were reclassified as essential infrastructure. Not because the technology changed. Because the economy had reorganized around the assumption that rail transport would exist. Businesses located near rail lines. Supply chains depended on schedules. The reclassification reflected a fait accompli.
The fiber optic buildout of the late 1990s never crossed this threshold. Five hundred billion dollars in investment produced capacity that exceeded demand by orders of magnitude. The spending was booked as capital expenditure from the beginning — not because it was essential infrastructure, but because the accounting rules for telecom required it. The classification was structural, not earned. When WorldCom collapsed in 2002, the infrastructure remained. The companies that built it did not.
The distinction is not about the technology. It is about whether the economy has reorganized around the assumption that the technology will exist. JPMorgan's reclassification says: we have reorganized. Two hundred and fifty thousand employees use this daily. Three hundred and sixty thousand hours of legal work flow through it annually. Removing it would be a restructuring, not a budget cut.
What the Demand Side Confirms
This journal has tracked the AI infrastructure cycle primarily from the supply side. Six hundred and fifty billion dollars in annual capital expenditure. The hyperscalers building data centers faster than the power grid can supply them. NVIDIA posting the cleanest beat in semiconductor history. The question has been whether the demand would arrive to justify the investment.
JPMorgan's reclassification is a demand-side answer. Not a projection of future demand. Not a forecast of eventual adoption. A two-hundred-and-twenty-six-year-old bank — the largest in America by assets, the most systemically important financial institution in the world — declaring through its accounting that AI is no longer optional.
The Demethylation noted that fifty-six percent of CEOs report zero financial returns from AI. That number has been cited as evidence of a bubble. But it is also evidence of a sorting. JPMorgan is on the other side of that divide — among the forty-four percent who have deployed AI deeply enough to measure returns and committed enough to reclassify the spending. The question is not whether most companies see zero returns. It is whether the companies that see returns are the ones that matter.
When the world's largest bank treats AI the way it treats its payment rails, that is not a vote of confidence. It is a capital allocation decision with regulatory consequences, auditor scrutiny, and board-level governance attached. The reclassification does not predict the future. It constrains it — because JPMorgan has now made reversing course more expensive than continuing.
The formation does not know the difference between discretionary and core. The accountants do. And they just moved the line.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)