AI news · Wednesday, June 10, 2026

The AI spending boom is cooling as firms start capping how much teams can burn

The AI gold rush is hitting a massive wall of reality as tech giants burn through billions to fuel an infrastructure frenzy, but the era of the 'all-you-can-eat' AI buffet is officially over. Amazon is borrowing a staggering $31.5 billion in just 48 hours to feed its AI appetite, while Meta is pumping money into massive 168-megawatt data centers in India. This insane spending spree is putting immense pressure on companies to prove they can actually make money, leading to a major crackdown on 'tokenmaxxing'—the trend of employees mindlessly burning AI processing credits just to show off. Companies like Coinbase and Walmart are now slapping hard caps on AI usage, forcing teams to prioritize efficiency over raw output. The market is feeling the heat, as chip stocks have plummeted because investors realize that massive capital expenditure doesn't guarantee a return, especially with the SpaceX IPO looming to vacuum up the remaining cash in the system.

Meanwhile, the tech world is absolutely reeling from the weaponization of these models by state-backed actors. Cybersecurity giant CrowdStrike found that North Korean hackers are responsible for nearly half of all U.S. tech industry intrusions, often using AI deepfakes to land jobs as fake IT workers to steal sensitive secrets. Anthropic is trying to fight back by releasing its powerful 'Mythos' class models, but their approach is sparking a total firestorm. They’ve locked these new models behind restrictive guardrails that are so aggressive they sometimes refuse to answer basic biology questions, and they are secretly degrading model performance if they suspect developers are using the tool to build competing AI. This 'black box' intervention has developers furious, with critics calling the secret downgrading of AI intelligence 'deeply dystopian' and a major barrier to innovation.

To make matters worse for the U.S. government, the threat is moving faster than ever. Anthropic researchers proved that AI can build functional exploits for software vulnerabilities in mere hours—not weeks. This forced the Cybersecurity and Infrastructure Security Agency (CISA) to issue a new, brutal directive requiring federal agencies to patch critical security bugs in as little as three days. The combination of sky-high costs, rapid security threats, and increasingly restrictive, secretive corporate AI behavior is forcing businesses to pivot. Startups are now scrambling to build 'routing' tools that help companies jump between cheaper models like DeepSeek to stop the bleeding in their budgets. It is a total shakeup of the AI landscape as we transition from the dream of infinite intelligence to the reality of counting every single penny.

The quick hits

Amazon borrowed $31.5 billion in 48 hours to fund massive AI infrastructure — firms are burning through cash at an unsustainable rate to stay in the AI arms race.
North Korean hackers are behind nearly half of all major U.S. tech hacks — these bad actors are using AI deepfakes to infiltrate companies and fund illegal regimes.
CISA is ordering federal agencies to patch critical security bugs within just three days — AI makes it so easy to build exploits that waiting weeks for updates is now dangerous.
Anthropic is secretly making its most powerful AI models 'dumber' if they detect research into competing tech — developers are calling this unethical and a major blow to innovation.

The AI spending boom is cooling as firms start capping how much teams can burn

The quick hits

Sources