Thoughts on AI in chemical distribution: what's working, what to skip, and how operators are putting it to use.
Most companies with a Claude license never train their people on it, so the tool sits unused. I offer training and hands-on workshops built on chemical data so teams get a return from day one.
Anthropic's annual run rate went from $1B to $30B in 15 months, and eight of the Fortune 10 are now Claude customers, even though most operators still picture ChatGPT when they think "AI." Brand recognition is a lousy proxy for which model is actually built for your work.
Anthropic signed a 300 MW compute deal with SpaceX for power from its Colossus 1 data center in Memphis, enough for roughly 250,000 homes. The AI race has shifted from who has the smartest model to who can plug in, and that changes what to watch when picking a vendor.
A friend sent me a chart of 60 AI tools across 12 categories, and most companies I work with are piloting a few without ever getting one fully working. My advice is to pick one tool, tie it to one costly workflow, and give it 90 days before adding more.
SpaceX is now selling AI compute from two massive new data centers, and Anthropic is one of its first big customers. When a company known for landing rockets on barges gets into compute, it shows how serious, and expensive, the AI infrastructure race has become.
Google switched Gemini to compute-based usage limits, matching what Anthropic already does with token consumption on Claude's Pro and Max plans. Claude Code and Cowork have been underpriced for what they deliver, and Google's move (matching Anthropic's model, without publishing the formula) probably buys Anthropic more time before they have to raise the price.
Datacor's Winter 2026 release reads a customer PDF and writes the sales order straight into the ERP, no template or OCR required. That's the silicon workforce arriving at distributors, an agent doing a job a person used to do.
SEO gets you ranked in Google's results. GEO gets you cited inside the answer ChatGPT, Google AI Overviews, or Copilot writes back to the user, and most sites are only built for the first one.
Builder.ai raised over $450 million pitching AI that built apps, when the work was actually done by about 700 engineers in India writing code by hand and told to work UK hours so it looked automated. It's still the best test for any AI vendor pitch, ask exactly where the model decides, what data trained it, and how you validate the results.
A tour of an Amazon fulfillment center got me asking why distributors still move people toward product when the product could move instead. Hazmat keeps robots out of our warehouses, and there are still lessons in how Amazon rebuilt the question from scratch.
A quote worth sitting with on the GM, Ford, and Stellantis job cuts. Cut your workforce beyond what AI can actually replace and you lose productivity, institutional knowledge, and your best people. Fail to rethink how work gets done and your competitors pass you by.
Setting up an AI project with your own price list, sales report, and SDS files takes about 10 minutes, and that's the step most companies skip. The technology is ready, the real gap is imagination for how to use it.
