Anthropic Eyes $1.4 Trillion IPO With a Factory-Floor Safety Playbook

Anthropic is pitching institutional investors ahead of a Nasdaq IPO expected next week under the ticker ANT, with bankers weighing a valuation that could top $1.4 trillion. To justify that number, the Claude maker has to prove cautious, traditional industries will trust its software with live company records.

Its latest proof point landed on the factory floor. Anthropic launched Claude for Manufacturing alongside a major partnership with industrial software giant Infor. The stakes are concrete: a 2025 Manufacturing Leadership Council survey found 79 percent of manufacturers still type data by hand between tools that do not connect.

The plugin ships with 17 pre-built skills and four connectors — to Autodesk Fusion, Infor, Microsoft 365, and Oracle NetSuite. That's the product. The governance model around it is the part worth watching.

What does the rollout actually look like?

Picture a real headache from the rollout: a steel shipment slips by two weeks. Normally, a buyer burns half the morning digging through spreadsheets and emails to find which customer orders are ruined. With the Infor setup, the buyer asks Claude in plain English which orders are hit and how much spare steel other plant sites hold. Once the buyer clicks approve, Claude releases a new purchase order to a backup supplier and reschedules production inside Infor.

Anthropic and Infor didn't plug a chatbot into a factory database and cross their fingers. They built a three-part safety playbook.

Rule 1: Match every AI action to a human badge

A software agent should never carry more power in the building than the person typing the prompt. The companies chained Claude directly to a worker's existing single sign-on and role-based permissions inside Infor. When a buyer, planner, or service rep asks a question, the software only sees and changes what that employee is already cleared to touch — and every move gets logged.

Infor also taught Claude factory slang upfront, so "item," "SKU," and "product" mean the same thing without workers explaining vocabulary one by one.

"When AI begins managing your workflows, the system underneath matters more, not less," Infor CEO Kevin Samuelson said.

Rule 2: Freeze the workflow at human review

Lauren Collett, head of enterprise industries go-to-market at Anthropic, noted workers lose huge chunks of their day moving information between disconnected systems. The 17 skills handle that scavenger hunt, then stop right before the finish line.

An estimator can have Claude read a customer request, flag missing details, check past job costs in Oracle NetSuite, and draft a quote. A technician facing a broken machine can feed Claude sensor exports, repair notes, photos, and manuals to rank likely culprits. Small shops on legacy systems can drop file exports into shared Microsoft 365 folders instead.

Then the machine stops. Before a quote goes to a client, a human signs off — and Claude links every figure back to its source document, so staff check the receipts in seconds. Jessica Vincent, vice president of business solutions at Oldcastle APG, said fast answers inside governed systems help employees make smart calls without putting company rules at risk.

Rule 3: Keep AI off the machinery

Claude can draft supplier requests, compare 3D designs in Autodesk Fusion, and shift production schedules — but it does not operate physical equipment. Instead, Anthropic is developing a separate research preview, the Model Hardware Standard, giving programmable machines a strict way to state their measurements, settings, and safety limits.

The takeaway for operators is blunt. Let AI handle the 79 percent of the day spent copying numbers between tabs, but lock its permissions to existing staff logins, demand source links on every draft, and keep a human finger on the final approve button. Whether Wall Street prices that caution at $1.4 trillion is the question next week's listing will answer.