AI, chips, and model wars tonight

A Hacker News thread titled “AWS: Inaccurate Estimated Billing Data – $1.7 billion” indicates strong community attention to the issue.

TechCrunch reports Amazon is fixing a bug that billed some AWS customers billions of dollars.

Techmeme says the billing failure is tied to an issue with unit pricing in the estimated billing computation subsystem.

Another related report references estimated bills reaching as high as $1.5T.

For teams using AWS heavily, this turns estimated-cost monitoring into a production safety concern.

Apple filed a trade secrets lawsuit against OpenAI last Friday, and it’s not messing around. The complaint alleges a pattern of misconduct reaching all the way up to OpenAI’s chief hardware officer and claims more than 400 former Apple employees now work at the company.

On this episode of TechCrunch’s Equity podcast, hosts Kirsten Korosec, Anthony Ha, and Sean O’Kane dig into what the lawsuit could mean for OpenAI’s own hardware ambitions and IPO timeline, plus a bigger theme running through the week’s news: how much should anyone trust AI companies with their data?

Most Popular Tesla driver in fatal Texas crash pressed accelerator 100%, NTSB confirms Sean O'Kane

Amid hardware legal battle, OpenAI releases a $230 keyboard for Codex Lucas Ropek

OpenAI’s first hardware device is reportedly a screenless speaker that can move Lucas Ropek

Techmeme reports that OpenRouter has discussed a possible sale to a larger technology company.

The reported valuation discussion is in the billions, above the $1.3B mark from May.

OpenRouter is presented as an abstraction layer helping developers access hundreds of AI models.

That means any ownership or strategic shift could affect routing behavior and cost surfaces.

Development teams should treat this as a potential change-control point in model orchestration.

Valar Atomics, a startup building small modular nuclear reactors (SMRs) — essentially miniaturized, factory-built power plants designed to be cheaper and faster to deploy than traditional reactors — is in talks to raise a new round of capital, according to three sources familiar with the company.

The Information was the first to report the funding discussions, including that the El Segundo, California, startup is raising a $1 billion equity round.

Part of that capital has been raised previously at a lower valuation, the people told TechCrunch. Specifically, Valar has raised $450 million — including $340 million in equity and $110 million in debt — at a $2 billion valuation, per a Bloomberg report in March.

Deals structured in multiple installments at varying valuations, occasionally executed at different times, are becoming increasingly common in today’s AI-fueled fundraising environment. These deals can create the perception that capital was invested at a single, uniform valuation.

Earlier this month, the company showed that its nuclear reactor provided a small amount of power to an Nvidia AI chip. Concurrently with that proof-of-concept demonstration, Valar and Nvidia announced a partnership to explore the development of nuclear energy to power future AI data centers.

Reuters reporting surfaced through Techmeme says the Philadelphia Semiconductor Index dropped about 10% over one week.

That is described as the index’s largest weekly fall in over a year.

It now sits about 20% below the late-June all-time high.

The same names that powered this year’s market rally are included in this correction.

For AI teams, this reinforces pressure on chip lead times and hardware budget predictability.