Apple goes thermonuclear on OpenAI, Musk and Altman trade jabs, and a subsidy price war rages between the labs — but NLW's read is that all this escalating acrimony is, at least for now, a gift to the rest of us. The competition has moved beyond frontier models to the entire ecosystem, and nobody has a handle on what shifts next.
AI competition just got personal — and the chaos is good for you.
The AI race is no longer just about who has the best frontier model. It's about hardware, architecture, learning loops, and geopolitics, all colliding in a liminal in-between moment where nobody knows where value will accrue. That anxiety is driving public spats — Apple's lawsuit, the Musk-Altman feud — but it's also driving a subsidy price war between the labs. NLW's takeaway: take advantage of the competitive upheaval while it lasts, because a smart individual can benefit mightily.
Apple filed a blockbuster suit alleging OpenAI stole hardware designs and IP, claiming 400+ former Apple employees joined OpenAI's hardware push. Centered on engineer Chang Liu — who allegedly used a software bug to access Apple servers ('I found out I can access the network storage. So funny') — the suit alleges OpenAI actively encouraged new hires to bring confidential parts to show-and-tell sessions.
The suit echoes Steve Jobs' 2010 vow of thermonuclear war on Android — with Tim Cook, as a final act as CEO, taking a similar approach to stifle OpenAI's device. Experts are split: California courts reject non-competes, so trade-secrets law is 'the only legal perimeter left around institutional knowledge,' and Apple pleaded squarely inside it.
NLW wants to see OpenAI's response before making up his mind, but agrees the lawsuit signals the core theme: it's no longer just about models, it's about the entire ecosystem around them. The alleged IP theft still seems 'strange and desperate' in a way that doesn't comport with OpenAI's self-image.
— Sam Altman. Musk retweeted the lawsuit ('They sure put a lot of effort into this crime') and resumed his feud with Altman. NLW crowned Altman the winner of the tit-for-tat, though Altman also cooled it on Apple: 'I have tremendous respect for them. S-tier company.'
Power users hit usage limits fast on the new model — partly because you now need to control reasoning effort rather than max everything, partly configuration issues on OpenAI's end. OpenAI temporarily removed the five-hour limit for Plus/Business/Pro, pushed efficiency changes, and reset usage after hitting 6M active users: 'Go do things.'
— Developer ECAS. Instead of poaching Anthropic users as GPT-5.6 launched, Anthropic extended its Fable trial twice and kept Claude Code limits 50% higher — turning the moment into a subsidy price war that benefits builders.
Semi Analysis found the $20/mo tier still allows ~$400 of Anthropic usage or $700 of OpenAI usage; the $200/mo tier runs to $8,000 (Anthropic) or a staggering $14,000 (OpenAI) in max tokens. NLW's advice: take advantage while it lasts, because the frontier won't be subsidized forever.
— Zhipu / Z.AI founder. Z.AI's founder defended open models, releasing GLM 5.2 (1M-token context) under the permissive MIT license: 'The heights we reach belong to all humanity, and the roads we build belong to everyone.' The company's CEO conceded they're not yet at Mythos level but expect to be by year's end.
— Satya Nadella, Microsoft CEO. Satya Nadella argued models learn from 'exhaust' — prompts, tool use, and especially corrections — which leaks 'trace by trace' into the model provider. His pitch: distribute learning infrastructure to every firm so they control their own learning loop. 'In consuming intelligence, you are creating intelligence, and what you create should belong to you.'
— Guillermo Rauch, Vercel CEO. Nadella cited Palantir's Alex Karp — customers 'want to know they own the means of production' — and Vercel's Guillermo Rauch summed up the enterprise playbook: own your data, evals, model choices, and software layer. Don't outsource your brain.
CNBC's Deirdre Bosa framed the shift NLW has charted for months, and even Big Short's Michael Burry retweeted it. For Burry it's how the bubble bursts; for Gavin Baker it's the 'mega bull case' — margin dollars redistributing from frontier labs (90%+ inference margins) to infra providers with the lowest per-token cost.
NLW pushes back on the assumption that leading labs won't participate in the cheaper-model shift. GPT-5.6's cheaper versions — Terra and Luna — are 'basically better than GLM performance for lower than GLM costs.' He also thinks the pace of change is wildly overstated: most firms are still trying to get employees to use Claude at all.
NLW's bottom line: the tectonic plates of AI competition are shifting, and nobody has a handle on what comes next — which is why the spats and acrimony are intensifying. But the subsidy extensions prove that, in the short term, this fierce competition is genuinely good for all of us.
Politico reports (citing nine people familiar) that the White House is in early-stage discussions about an EO to deal with the perceived security threat of Chinese open-source AI. Officials denied it, but the tea leaves point that direction — likely galvanized by headlines around GLM 5.2 rather than any real benchmark testing.
— Nathan Lambert, Interconnects. Interconnects' Nathan Lambert warned that open-source AI is 'staring down the barrel of policy action that could make open models a permanent second-class citizen,' as advocates grow wary of impending US restrictions.
Commerce ruled that the UAE government and approved firms can access advanced AI chips without a license, paving the way for Gulf mega-clusters. Last year's deal mentioned 500,000 chips, but once G42 and MGX are approved there's effectively no limit — a first for a non-NATO, non-treaty nation.
— Chris Maguire, Council on Foreign Relations. China hawk Chris Maguire warned the world's largest data centers will now sit in the UAE, operated by firms that could provide backdoor access to China. Elizabeth Warren called the deal corrupt over Trump's family crypto ties.
— Ryan Fetesiyuk, American Enterprise Institute. AEI's Ryan Fetesiyuk argued the concerns are overblown given where the world is heading: the Gulf is becoming an essential node in a globally distributed network of US AI inference hubs, and the US should install 'chips in sockets' fast while China's chipmaking is still immature.
The Korean memory maker raised $26.5B in its Nasdaq debut, edging out Alibaba's 2014 listing, and popped 13% on day one — a strong result amid a 9% semiconductor correction this month.
— Chey Tae-Won, SK Hynix Chairman. SK Hynix Chairman Chey Tae-Won dismissed oversupply fears, saying customers keep telling him doubling capacity 'is not enough.' Demand from AI agents and physical robots, he argued, is 'enormous, exponential.'
Meta's new image model let users generate images of others by tagging their Instagram accounts, pulling public photos as context. After the Screen Actors Guild urged members to disable it, Meta pulled it — another marker of where the public draws the line on AI imagery.