Burgundy
AI

Who Pays for AI

On the eve of the US midterms, the real fight over artificial intelligence is about its bills, how far it runs on its own, and whether anyone trusts it.

Burgundy
3 min read ·

On the eve of the US midterms, the biggest question over artificial intelligence is a dull one: who pays for it. The fight over AI data centers has gone bipartisan, the research newsletter SemiAnalysis wrote on November 1, and a long permit freeze in Texas is now the single biggest risk to America's 2028 buildout. Most of the 62.2 GW of net US additions projected for 2028 is not yet under construction, SemiAnalysis estimates, unlike 2027's capacity. Thirteen states hold about 70% of the expected 2028 capacity. Ten of them elect governors on November 3.

What those candidates want is telling. Of 20 nominees SemiAnalysis analysed, nine back a construction pause, and 15 want data centers to pay their own grid costs. Texas, roughly 30% of 2028 additions, froze permits on September 21, and whether that freeze lifts may hinge on the governor's race. OpenAI, meanwhile, is weighing a post-midterm push that could rename data centers 'AI compute campuses', TechCrunch reported via Axios. One line caught the strategy: 'Data centers are deeply unpopular. But what if we called them something else?'

The money question runs through earnings season too. Last quarter the gap between Meta's ad-revenue growth and Google's search growth was the widest since 2021, CNBC reported, arguing that the same AI shift is boosting Meta's ads while hurting Google search. Santander, Novartis and others told investors AI is starting to lift revenue, cut costs or improve profitability. At the same time, Amazon and JPMorgan are among the companies blocking personal AI agents such as Meta's Muse, though retail experts told CNBC that friction-removing technology wins out over time.

China is racing on another front: autonomy. The start-up DeepSeek previewed a 'super agent' meant to run multiple tasks with minimal human oversight, cutting the work of coordinating several AI agents, the South China Morning Post reported. Z.ai founder Tang Jie told an audience in Shanghai that using AI to build smarter AI makes artificial general intelligence more achievable, though he said it has to be done safely. The Post says labs in both countries are chasing that recursive self-improvement.

Trust was the night's other thread. Security firm Parameter Labs chained bugs in ChatGPT's HTML previews, a math library's virtual keyboard and YouTube's oEmbed endpoint, so that a crafted share link could steal a signed-in visitor's data or act as them, Simon Willison reported. OpenAI has fixed it and paid a $5,000 bounty, which Willison called cheap for a flaw this serious.

Elsewhere, Jepsen's Kyle Kingsbury wrote that LLMs erode trust in documentation and software, because the marks of a mature project are now easy to counterfeit. IEEE Spectrum's Harry Goldstein said about half of a recent scroll through his feed was AI slop. Pitjantjatjara artist Sally Scales, who chairs the National Gallery of Australia's First Nations advisory group, said AI is turning Indigenous art into 'Temu slop' and called it 'a theft'. And Wired reported that chatbots are leading hikers astray, with potentially deadly results, when people hand over their planning and their judgment.

Sources

  1. The Datacenter Backlash Goes Bipartisan: Next Week's Midterms and the 2028 Buildout · SemiAnalysis
  2. OpenAI might rebrand its data centers as ‘AI compute campuses’ · TechCrunch, AI
  3. Why AI is boosting Meta's ad business while hurting Google search · CNBC, Technology
  4. Companies promised AI gains, here's how the payoff showed up in earnings · CNBC, Technology
  5. For JPMorgan, Amazon, fighting AI bots like Meta Muse may not be a winning long-term strategy · CNBC, Technology
  6. DeepSeek unveils AI ‘super agent’ in coding race with OpenAI, Anthropic · South China Morning Post, Tech
  7. Self-improving AI the key step towards AGI but safety is key: Z.ai founder · South China Morning Post, Tech
  8. How We Took Over ChatGPT Accounts · Simon Willison's Weblog
  9. Quoting Kyle Kingsbury · Simon Willison's Weblog
  10. Guarding the Gates of Knowledge · IEEE Spectrum, Artificial Intelligence
  11. AI is stealing Indigenous Australian art and churning out ‘Temu slop’, artist says · The Guardian, Artificial intelligence (AI)
  12. Chatbots Are Leading Hikers Astray, With Dangerous Consequences · Wired, AI

More from Burgundy