As a reminder, here are the five points I wanted to make. Part 1 handled (1) and (2):
AI is in some ways a useful tool. It’s usefulness is way overhyped, but critics also overreact and misunderstand it.
Much of the use of AI we see today is due to it being pushed onto corporations and individuals. The true use of AI will be seen when the push is done.
Because of the overhyping and the pushing, the demand for AI is way overestimated, so we’re in a datacenter bubble. Any and all efforts to slow down or stop datacenter building are good.
When the bubble pops, a bunch of banks and corporations will be looking for a bailout. We’ve got to be ready to stop them.
Some AI is based on theft of intellectual property and there needs to be a reckoning.
Demand is Overestimated
Like many other new technologies, AI vendors have basically given away their technology to drive adoption. This is still happening in the consumer space, but businesses are starting to pay, and when they do, they are shocked at how much AI costs. Because AI is billed in “tokens”, this has been called “Tokenpocalypse”:
"We're seeing from some of the data internally at least that it's actually not our engineers that are driving the token consumption. It's a lot of the non-engineers that are doing some of those behaviors [...] you were talking about," Accenture's agentic AI strategy lead Justice Kwak said in the meeting, according to the leaked audio.
The pattern may not be limited to Accenture. Some providers, including GitHub, are reportedly shifting away from flat subscriptions and toward per-token pricing, while Uber is said to have capped workers' use of AI coding tools after previously urging staff to "use AI as much as possible."
For businesses, there’s also the issue of both Google and Microsoft adding their AI to their business products (Google Workspace, which is Gmail for companies, and Outlook). In both those cases, AI is essentially forced on the user. Google is especially bad about this, summarizing simple emails and generating AI replies without being asked to do so. If Google charged for this service, I’ll guarantee you that I, for one, would figure out how to shut it off. It provides negative value to me, since it takes up screen space.
At a higher level, there’s a hype cycle going on where Anthropic and Open AI, the two market leaders, are pouring their money (from investors and banks) into cloud computing providers Google, Microsoft and Amazon. This allows the big traditional tech companies to claim that they’ve got big “AI” business when in fact most of it is the AI companies buying computing resources from them. Since Open AI and Anthropic are spectacularly unprofitable, at some point this merry-go-round is going to stop. Commenter Pacem Appellant linked to the journalism of Ed Zitron in the last post (His YouTube channel is also worth watching). Here’s Zitron’s basic point on circular financing:
The Future Growth of Google, Microsoft, and Amazon Is Contingent On Anthropic and OpenAI Spending $200bn+ in 2027, Which Requires $250bn to $300bn in Funding
[..]
Right now, there are (outside of hyperscalers buying compute for them, and whatever it is Meta is up to) two companies that spend more than $500 million a year on AI compute, namely Anthropic and OpenAI. Both are unprofitable, and both lose tens of billions of dollars a year.
If this is true, or even close to true, there’s going to be a some point where AI funding is going to stop. This means that the current datacenter construction will stop, which means that we’ll have a huge amount of empty, half-built, or unused data centers.
Politically, this means that everything we can do to slow or stop data center development is essential to lessen the devastation to small communities which won’t get the tax benefits and (admittedly few) jobs they were promised. This is where I’d support an “all of the above” strategy. This includes both moratoriums and a list of reforms similar to James Talarico’s:
setting minimum federal requirements to protect the grid, environment, jobs and transparency related to new data center development, “preventing a race to the bottom for communities competing to recruit data centers;”
ending “sweetheart tax deals for big tech companies” by repealing the state’s data center sales tax exemption;
ensuring large energy users pay for their own infrastructure and grid interconnection costs;
requiring data centers use closed-loop water systems that reuse a fixed amount of water;
granting local communities the authority to approve and reject data center development in their towns;
releasing government records detailing individual tax break data for data center facilities in Texas.
Trump will never sign a bill with this in it, unless it’s part of some leverage to pass a budget, but states can do this.
The other thing we can do is to connect the higher price of phones, computers and any other device to data centers hogging up all the memory — the name for this is RAMageddon.
No Bailouts When the Bubble Pops
Scott sent in this good piece at the American Prospect by the Revolving Door Project, which details the failures of the Obama Administration in the bank bailout, and encourages Democrats to not fall for it again with AI. Banks and billionaires like Musk, Ellison and others are going to come running and begging the minute that the AI bubble begins to burst. There’s no reason to bail these fools out. They can lose their money that they bet foolishly. Microsoft, Amazon and Google can afford to write down their AI investments. Because a good part of the Democratic Party has been captured by AI donors, this might be difficult. However, the 2026 elections have shown that incumbents can lose, and they can be knocked out by challengers who don’t take corporate donations and make them an issue in the primaries. I’m guessing 2028 will be worse for donor-captured incumbents.
Intellectual Property Theft
If there is some kind of bailout of AI companies, it needs to be contingent on those companies paying the creators whose work they stole. AI models have been trained on books, art and music that aren’t in the public domain, one of the largest thefts of intellectual property in our history. The AI companies either need to purge those works from their models and re-train them (expensive) or go through their training materials and being paying royalties to all the authors whose works were stolen (expensive and complicated). If this is done correctly, I’m guessing that those AI companies will go for door #1: a purge. Then they can start licensing work from artists, like your local bar, bowling alley or movie theater has been doing for almost 100 years.
One point on this: for business use of AI, a lot of the training materials are things that their authors want the AI models to use. Salesforce and Adobe, to pick two of many examples, should be happy to let AI train from their online documentation. This will allow business users to query AI to answer questions about big, complicated software products. The intellectual property issues around AI are mostly about consumer use of AI, not use by programmers or business analysts.
If I can distill all of this down to one sentence, it’s this: You aren’t a “luddite” or “technophobe” if you oppose data centers — you’re someone who’s looked at the facts and decided not to participate in another bubble.
