The Register's Rupert Goodwins reports that someone has released source code and binaries for AI-generated, clean-room Rust reimplementations of Microsoft Office and Adobe Creative Suite. They are pre-alpha and buggy but substantially functional, raising legal questions because functionality cannot be copyrighted. Goodwins notes the irony of companies that trained AI on what he calls the biggest IP theft in history now claiming AI shouldn't sully their own products.
This is part of a broader pattern where AI disrupts its creators. Internal Microsoft-OpenAI emails reportedly acknowledge a "doom loop": AI siphons away the ad revenue that pays for content, drying up the data it depends on. Meanwhile, the capital call for AI infrastructure is enormous. Gartner says there is not enough return on investment to drive growth, while SemiAnalysis projects $11 trillion in cumulative capex by 2030 and asks where the money will come from.
The investment structure is unusual. Neocloud startups lack investment-grade status, so Nvidia and other cash-rich firms backstop them by guaranteeing to rent unsold capacity, shifting risk onto rated balance sheets. Older datacenter chips are also not depreciating as expected, adding strange, hard-to-model factors. The column concludes that nobody knows how the capex will be funded.