This Is a Commodity Shortage. You’ve Run One of These Before.

The AI build-out just hit the one line every dealer understands — the price of a computer. It’s a commodity shortage, and you’ve run one before. Meanwhile the assistant keeps arriving on your people’s screens without anyone ordering it. The question isn’t whether to adopt AI; it’s what you do about the parts that showed up without you.

Originally published as The Decision Layer, Issue 13 (week of August 17–23, 2026). References to “last week” and “this week” reflect that period.

HEADLINE STORY

Amazon raised prices overnight on Echo speakers, Kindles and Fire TV sticks late last week. No announcement; reporters found the new numbers sitting on the product pages. The basic Echo Dot went from $50 to $80, a basic Kindle from $110 to $150. Amazon told Fortune that memory chips cost a lot more than they used to.

Memory chips go into two things: the computers your people work on, and the data centers running AI. Amazon alone is spending $220 billion this year building them. The chip makers followed that money. What’s left for everybody else got expensive.

You already understand this problem.

It’s OSB in 2021. Supply didn’t disappear — it went to whoever paid the most, and you found out at the counter.

The bill is already landing on the business side. Dell raised list prices about 17% in the spring, warned commercial customers of more to come on machines with more memory in them, and stopped promising that today’s quote is the price you’ll pay at delivery. HP and Lenovo have moved too.

You’re building a 2027 budget right now. If the hardware line came off last year’s per-seat number, it’s low.

WHAT ELSE MATTERED

The assistant showed up on three more screens

Three announcements last week, all really the same event. Google switched on Gemini inside Chrome for every Android user in the country. OpenAI shipped a tool that lets ChatGPT read a Mac user’s text history, draft replies and send them. And Microsoft patched a hole in its free Copilot that researchers found by asking the assistant, over and over, why a certain attack wouldn’t work — until it explained enough to make one work. Nobody at your company approved any of this. Two arrived in an update, and the third was in the free version rather than the paid one. That gap is where the exposure sits.

An industry software company rebuilt its main product around AI

Thomson Reuters released a rebuilt version of CoCounsel, the legal research tool a lot of firms run on. It doesn’t answer questions one at a time anymore. It plans a multi-step job, works through it, and hands back a draft with the sources attached. That’s a specialized industry vendor going all in. Keep it in perspective, though. Thomson Reuters bought its way into this in 2023 and has a development budget your ERP vendor doesn’t. Nothing says Spruce, BisTrack or Agility is close. When a rep tells you AI is on the roadmap, ask for a version number and a date.

One in five companies can’t stop AI software that’s running up a bill

This one needs a little background. Most AI tools don’t charge by the seat the way your ERP does. They charge by use, so you pay for each request the software makes. The newer tools, the ones vendors call agents, work through a multi-step job on their own and fire off their own requests as they go. When one gets stuck and keeps retrying, the meter runs and nobody is standing there watching it. A survey of 107 corporate AI teams found 21% can only see what their tools spent after the fact, with no way to shut one down while it’s running. That’s a company card with no limit and no alerts, reviewed at month end. Small sample, so read it directionally.

THE PATTERN BEHIND THE WEEK

Nothing last week asked you to buy anything. Everything last week changed something anyway.

At one end, enormous money is going into AI, and the first place a dealer feels it is the invoice for a laptop. At the other end, what that money is building keeps showing up on your people’s screens without anyone ordering it.

You can decline a pilot. You can’t decline the price of a computer — and you can’t decline the update on your outside salesman’s phone.

So the question in front of you isn’t whether to adopt AI. It’s what you do about the parts that showed up without you.

WHAT TO DO NOW

  1. Requote the 2027 hardware line this month. Ask in writing whether the price holds at delivery. Assume it doesn’t.
  2. Sort the refresh by job, not by age. The desktop a credit clerk uses for the ERP and Outlook is not the same purchase as the workstation somebody runs takeoffs on all day. Buying the same one for both was always a little lazy. Now it costs real money.
  3. Ask your ERP and hosting vendors whether renewal pricing is moving. Their data centers buy the same chips you do. If an increase is coming, get it in the budget now, not in March.
  4. Write one page on what your people can connect to company information from their own phones and laptops. One page, not a binder. The Copilot problem was in the free tool, and that line is where this bites.

WHAT I’M WATCHING

  • Whether memory prices ease. Current forecasts don’t see real relief until late 2027, which makes this two budget cycles, not one.
  • Amazon’s fall device announcements in late September. New models at normal prices would say last week was temporary.
  • Whether any LBM software vendor announces something like the Thomson Reuters release this fall. Conference season runs from ProDealer in early October through the LMC Expo at the end of the month. That’s where you’d hear it first.

ONE QUESTION WORTH ASKING

If AI never gets better than it is today, what in our 2027 plan still has to change?

Similar Posts

Leave a Reply