Four things, every day.

Hey. Meta, Microsoft, Nvidia, and 20 others just told Washington to leave free AI models alone. Nvidia signed a $500 billion deal in South Korea. Samsung's chip workers are getting $340,000 bonuses this year, and the story behind that number is messier than the headline.

Meta, Microsoft, and Nvidia just defended free AI models.

Meta, Microsoft, Nvidia, IBM, Dell, Hugging Face, Mistral, and 17 others signed an open letter July 24 urging US policymakers to protect open-weight AI, arguing it lowers costs, increases competition, and prevents lock-in to a handful of closed providers, per AI News.

It's a direct response to Washington weighing curbs on open models after Kimi K3, and it lands the same week the White House accused Moonshot of distilling Anthropic's Fable model to train K3. The letter argues distillation itself is standard practice and shouldn't get banned to punish alleged misuse.

Chipmakers and cloud providers have an obvious reason to want open models everywhere: more deployable models sells more hardware, regardless of who trained them. Read the letter as industry positioning, not settled policy.

Nvidia just signed a $500 billion deal in South Korea.

Nvidia and SK Group signed letters of intent July 24 for a partnership worth more than $500 billion, including a 2-gigawatt AI data center from SK Telecom using Nvidia's Vera Rubin platform, and a long-term deal locking in SK Hynix's next-generation HBM4 memory supply, per Tom's Hardware.

SK Hynix controls 60-70% of leading-edge HBM capacity, the memory type every AI accelerator needs. Locking that supply doesn't just secure Nvidia's own chips, it makes it structurally harder for AMD or any custom-silicon rival to get the same memory at the same volume.

This is a supply moat, not just a factory announcement. The first data center goes live in 2027, but the memory lock-in starts working against competitors immediately.

Samsung's chip workers are getting $340,000 bonuses.

Samsung will distribute roughly $26.6 billion in bonuses to its 78,000 semiconductor employees this year, averaging $340,000 per worker, after a union deal that narrowly averted an 18-day strike. Some memory-division workers could see closer to $400,000, per Bloomberg. Samsung's 2026 operating profit is projected to grow sevenfold on AI memory demand.

The union leader who won this deal, Choi Seung-ho, was hailed as a hero in May. By July, Bloomberg reports the celebration had already turned, with shareholder lawsuits and internal division over how the payout was structured.

The AI boom is generating real worker income, not just corporate profit, but this story is a reminder that a big win doesn't always stay simple once the money actually arrives.

The open letter has one very specific carve-out.

Buried in the July 24 letter is a direct defense of distillation, training one model using another model's outputs. The signatories call it standard machine learning practice and argue it shouldn't be swept into restrictions meant to target theft.

That's not a neutral technical point right now. It's the exact accusation the White House made against Moonshot days earlier, that Kimi K3 was built by distilling Anthropic's Fable model through banned chips routed via Thailand. The letter draws a line between legitimate distillation and "unlawful efforts to extract value," without saying which side any specific case falls on.

Watch where this line actually gets drawn. It will decide whether the next fast-growing open model gets treated as innovation or theft.

That's the four things today. See you tomorrow.