# The Deployment Shift — Enterprise AI, May–July 2026

Published 2026-07-06 · by Artjoms Krasnopjorovs (https://krasnopjorovs.com/)

This morning my news pipeline flagged a pattern: the biggest AI companies are pouring billions into forward-deployed engineering.

Old habit: I fact-checked it. All four announcements are real, and together they're bigger than any single headline:

Early May — Anthropic: AI services company with Blackstone, Hellman & Friedman, Goldman Sachs.
May 11 — OpenAI Deployment Company: $4B+ initial investment, TPG-led, with Goldman Sachs, Bain Capital, McKinsey and others.
June 30 — AWS: $1B into a Forward Deployed Engineering org, "thousands" of FDEs.
July 2 — Microsoft Frontier Company: $2.5B, 6,000 industry and engineering experts embedded inside customers.

What strikes me: there's no consensus in this industry — there's a widening split.

On one side: labs and hyperscalers, spending billions to put their engineers inside customer organizations.

On the other: customers with confidential data who want the perimeter to stay theirs. Palantir's CEO channeled them on CNBC this week — enterprises are "livid," paying for tokens while labs "steal the weights and alpha" of their business. He's selling the on-prem alternative, of course; the anger he's channeling is real anyway.

Two camps. Opposite incentives. Same signal: the bottleneck is no longer capability — it's deployment. Both sides are now fighting over the same square meter of ground: inside the customer's perimeter.

Good-enough intelligence is commoditizing toward the price of electricity. What doesn't commoditize is getting it through messy data, legacy systems, and compliance sign-off into production — behind a firewall the customer actually controls.

The hardest version of that problem is regulated industries. Take banking: data moves only where regulators allow — approved rails, approved parties, approved jurisdictions — every workflow has an auditor, and "just call the API" stopped being an answer the day the API learned to say no.

That intersection — regulated software development — is where I work: 20+ years in tech, the last twelve on cross-border payment rails inside a global bank.

Local LLM deployment is what my past nine months were about: from first quantisation to the multi-node cluster that started this post.

Personally, I'm glad my university years on neural networks gave me a head start into the modern stack — and that two decades of corporate delivery taught me which roadblocks these systems hit before they ever reach production.

Deployment is the product. The model is a component. Sovereignty is the feature.

P.S. Yes — the pipeline that flagged this story is built by me, running locally, on open-source tools and open-weight models. The thesis demonstrates itself.

Source: https://krasnopjorovs.com/articles/the-deployment-shift-enterprise-ai-may-july-2026