Intelligence is foundation
Subscribe
  • Luma
  • About
  • Sources
  • Ecosystem
  • Nura
  • Marbl Codes
00:00
Contact
[email protected]
Connect
  • YouTube
  • LinkedIn
  • GitHub
Legal
Privacy Cookies Terms
  1. Home›
  2. Featured›
  3. Builders & Makers›
  4. JPMorgan Stopped Calling AI an Experiment
Builders & Makers Wednesday, 13 May 2026

JPMorgan Stopped Calling AI an Experiment

Share: LinkedIn
JPMorgan Stopped Calling AI an Experiment

JPMorgan reclassified AI investments from research and development to core infrastructure. That single accounting decision reveals more about where enterprise AI is heading than any product announcement this year.

Infrastructure isn't experimental. It doesn't have a trial period. It requires governance, accountability, disaster recovery plans, and someone who gets fired when it fails. R&D budgets tolerate failure. Infrastructure budgets don't. When JPMorgan made this shift, they weren't making a prediction about AI's future - they were acknowledging what it already is inside their operations. Core. Load-bearing. Non-optional.

The timing matters. Most enterprises are still treating AI as a pilot programme. Testing use cases. Running proofs of concept. Measuring ROI on isolated deployments. JPMorgan skipped that phase. They're not asking whether AI belongs in the infrastructure stack. They're asking how to operate it at the rigour level infrastructure demands. That's a different conversation entirely.

What Infrastructure Classification Actually Means

Reclassifying AI as infrastructure changes who owns it and how it's funded. R&D sits in innovation teams with experimental budgets. Infrastructure sits in operations with reliability requirements. Different accountability. Different expectations. Different consequences when something breaks.

Infrastructure needs redundancy. If your customer database goes down, you need a backup live within minutes. If your AI inference pipeline goes down, the same standard applies. That means duplicate systems, fallback models, monitoring at every layer. It means service level agreements with penalties. It means 3am pages when latency spikes. The operational burden is completely different from experimental AI.

Infrastructure needs governance. Who decides what data the model can access? Who approves changes to prompts that alter behaviour? Who audits outputs for compliance? These aren't product questions - they're policy questions. And they require frameworks most companies don't have. JPMorgan has compliance infrastructure for every other system. Now AI needs to fit into that same framework. That's why reclassification matters - it forces the governance question before deployment, not after failure.

The Data Pipeline Problem

The article identifies what most enterprises are missing: the data infrastructure to operate AI at scale. It's not about having data - every large company has enormous amounts of data. It's about having it organised, accessible, and clean enough to use reliably. Most enterprise data lives in silos. Legacy systems that don't talk to each other. Inconsistent formats. No unified access layer. Governance policies that vary by department.

AI makes this problem visible in a way previous technology waves didn't. A business intelligence dashboard can work with batch data updated overnight. An AI system answering customer queries needs real-time access to current information across multiple sources. The latency requirements are tighter. The integration complexity is higher. And the consequences of stale or incorrect data are immediately visible to users.

Building this infrastructure is expensive and slow. It's not a software purchase - it's data engineering work, often requiring years of migration and standardisation. Companies that invested in data platforms over the past decade have an advantage. Companies that didn't are discovering they can't deploy AI meaningfully until they solve the underlying data problem. The model is the easy part. The pipeline is where enterprises get stuck.

Why This Signals a Shift

JPMorgan isn't the first bank to deploy AI. They're the first major financial institution to publicly classify it as infrastructure. That distinction matters because infrastructure classification is a signal to the rest of the industry. When a regulated, risk-averse institution calls something core infrastructure, competitors take note. Not because JPMorgan is ahead technically - but because they're ahead operationally. They've answered the governance, reliability, and accountability questions. That gives them a template others can follow.

For builders, this shift changes where value accumulates. Less in model training, more in deployment frameworks. Less in API wrappers, more in integration layers that connect AI to existing enterprise systems. Less in experimentation, more in operational tooling - monitoring, observability, incident response. The infrastructure layer around AI is still immature. JPMorgan's reclassification creates demand for products that treat AI like infrastructure, not research.

The broader pattern is this: AI is moving from innovation teams to operations teams. From experiments to dependencies. From nice-to-have to load-bearing. That transition requires different skills, different tooling, and different accountability structures. Most enterprises aren't there yet. JPMorgan is. And when a bank leads an infrastructure shift, the rest of the enterprise world generally follows. Slowly, cautiously, but inevitably.

More Featured Insights

Robotics & Automation
The Microrobots That Navigate Without Software
Voices & Thought Leaders
Why OpenAI's Enterprise Push Looks More Like 1975 Than 2015

Video Sources

AI Engineer
Lessons from Trillion Token Deployments at Fortune 500s - Alessandro Cappelli, Adaptive ML
Theo (t3.gg)
Stop letting your agents write Markdown.
Google DeepMind
Reimagining a 50-year-old interface (the mouse pointer) with AI
AI Engineer
Give Your Agent a Computer - Nico Albanese, Vercel
Matthew Berman
What happened to Anthropic?

Today's Sources

DEV.to AI
When JPMorgan Calls AI "Core Infrastructure," the Rest of the Enterprise World Should Listen
Towards Data Science
From Vibe Coding to Spec-Driven Development
DEV.to AI
Understanding Synthetic Users and Synthetic Data: The Future of AI-Powered Market Research
Robohub
Developing active and flexible microrobots
Hackaday Robotics
The Dark Side of Unitree Robot Dogs
The Robot Report
Comau and OMRON Robotics partner to offer robotics for more industries
ROS Discourse
Parallel robots in ROS
ROS Discourse
QERRA-v2 Classical - Call for Early Testers
Ben Thompson Stratechery
The Deployment Company, Back to the 70s, Apple and Intel
Latent Space
[AINews] The End of Finetuning

About the Curator

Richard Bland
Richard Bland
Founder, Marbl Codes

27+ years in software development, curating the tech news that matters.

Subscribe RSS Feed
View Full Digest Today's Intelligence
Richard Bland
About Sources Privacy Cookies Terms Thou Art That
MEM Digital Ltd t/a Marbl Codes
Co. 13753194 (England & Wales)
VAT: 400325657
3-4 Brittens Court, Clifton Reynes, Olney, MK46 5LG
© 2026 MEM Digital Ltd