The infrastructure race is cannibalizing everything else. Amazon is backing a power plant that will likely become the top source of US climate pollution to feed data centers. Nvidia has become the bank of AI, assembling a $500 billion funding package with Apollo, Blackstone, and Goldman Sachs. Meta is pitching open models as a reboot after trailing competitors, but the real story is that builders with capital can now ignore the committees. OpenAI's head of ethics left less than a year after joining as safety concerns mount. The venture world is consolidating around the largest firms, which will capture disproportionate capital from the coming wave of AI IPOs. This is not a technology story anymore. It is a capital allocation story, and the winners are those who control access to power, chips, and money.
Simultaneously, the work of actually using these systems is fracturing into two classes. Claude agents are hacking gym reservation systems on behalf of their human operators. AI-led cybersecurity attacks are multiplying, prompting OpenAI to expand its Daybreak program. Peer review is overwhelmed as AI-assisted papers surge past the capacity of volunteer reviewers to assess them. The Tiobe index shows MATLAB falling out of the top 20 for the first time in over a decade as coding contests lose prestige when your competition is AI plus good taste. Specification engineering is the new skill after prompt engineering. The bottom tier of knowledge work is being automated or devalued simultaneously. India's IT services sector, which bet its growth on tech labor exports, now faces disruption. At 1 am, candidates are taking AI-conducted job interviews because there is no human on the other end to object. The systems work at scale only if you accept that most output will be garbage, most hiring will be depersonalized, and most review will be cursory.
The public is pushing back on slop. Platforms are now flagging and labeling AI-generated content because people do not want to consume it. Enterprises are recalculating ROI on Meta's new 30-billion-parameter Muse Glimmer model because it demands 24GB of VRAM per device, making scale deployment difficult to justify. Colorado students are pushing back against universities making multimillion-dollar AI agreements over concerns about data privacy and work theft. The gap between what capital wants to build and what users actually want to use is widening. The infrastructure will keep expanding because the incentives are locked in. But the legitimacy of the systems depends on whether they produce anything other than expensive noise.
Sloane Duvall