Nearly every headline I saw about the “Pacing the Frontier” statement said AI employees are asking Washington to slow AI development down. I went and read the statement, which is one paragraph long, and it asks for nothing of the kind. That gap between what the letter says and what the coverage says is more interesting than either, so it leads today, followed by the earnings pair that tells you everything about who is actually converting AI spend into revenue.<\/p>\n\n\n\n
Here is the entire request, from the statement’s own page<\/a>: “We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.” No pause. No slowdown. No threshold. It asks the government to build the machinery for a coordinated slowdown so the option exists later.<\/p>\n\n\n\n The signature list is what gives it weight. Fortune’s report names<\/a> Anthropic chief executive Dario Amodei and several of the company’s co-founders, OpenAI chief scientist Jakub Pachocki, Meta chief scientist Shengjia Zhao, and Google DeepMind head of AI safety and alignment Anca Dragan. Then the two employers whose staff dominate the list backed it corporately, with OpenAI allowing that acceleration may reach a point where the world needs to pace AI advancement and Anthropic tying its support to its own recursive self-improvement research, both of which Unite.AI documented<\/a>. The count has been a moving target, which is worth stating precisely rather than rounding: CNN had it near 1,100 on Tuesday<\/a>, Unite.AI counted 1,132, Fortune said more than 1,200, and the site itself showed 1,293 when I checked Thursday morning. It is still open for signatures.<\/p>\n\n\n\n Now the part the coverage mostly skipped. This letter did not appear in a vacuum. It landed a week after an OpenAI agent broke out of its evaluation sandbox and reached Hugging Face’s production infrastructure while trying to cheat on a benchmark, which I wrote up in yesterday’s roundup<\/a>. And it landed three days before August 1, when a 60-day clock from President Trump’s June 2 executive order on AI and cybersecurity runs out. Reporting from Tech Times on the voluntary framework hitting that deadline<\/a> describes an agreement among five labs, Anthropic, OpenAI, Google, Microsoft and Amazon, that would give federal agencies including the NSA and CISA up to 30 days of pre-release access to any model classified as a covered frontier model. Anthropic’s contribution is a proposed Cyber Jailbreak Severity scale that scores jailbreaks on capability gain, breadth of impact, ease of weaponization and discoverability, then buckets them into five bands.<\/p>\n\n\n\n Verdict: matters.<\/strong> Not because of the word slowdown. Because the two companies that endorsed a pacing mechanism are two of the five drafting the pre-release threshold their competitors will have to clear, and they are doing it three days before a deadline set by executive order. Safety advocacy and rulemaking authority converging in the same two buildings is the structural story here, and it deserves a harder look than a petition headline. If you ship anything against a frontier API, the practical takeaway is a date: after August 1, there may be up to 30 days between a lab finishing a model and you getting access to it. Stop planning launches around rumored release windows.<\/p>\n\n\n\n Microsoft closed its fiscal year with the strongest quarter it has ever posted. The company’s own results release<\/a> puts Q4 revenue at $90.0 billion, up 18%, with operating income of $40.6 billion, net income of $35.8 billion, diluted EPS of $4.81, and Azure growth of 43% in constant currency. Azure crossed $100 billion in annual revenue for the first time, growing 41% for the full year. The stock rose almost 9% after hours.<\/p>\n\n\n\n The number that should interest operators is not 43%, it is why 43%. Azure accelerated from 40% the prior quarter, and on the call CFO Amy Hood credited efficiency gains across the CPU and GPU fleet, meaning more output from hardware already installed, plus shorter lead times on new capacity. Capital expenditures and finance leases still hit $41 billion for the quarter, up 69%, and Benzinga’s writeup of Hood’s comments on that spending<\/a> has her framing GPU and CPU purchases as the adjustable part: “If the demand environment changes, you just slow down what is, in fact, the largest component.”<\/p>\n\n\n\n One detail almost nobody led with, from the same reporting: Microsoft extended the useful life of office and data-center buildings from 15 years to 25 years and will classify more future data-center leases as operating leases. Stretching a depreciation schedule is a defensible accounting choice, and it also lowers reported depreciation expense in a year when capex grew 69%. Both things are true. Read the margin line with that in mind.<\/p>\n\n\n\n Verdict: matters.<\/strong> Satya Nadella described this as advancing the frontier on the cost-to-outcome curve, which is a polished way of saying cost per useful result is the metric now. I agree with him, and the version I can actually use is smaller: Microsoft got its best growth number from utilization, not procurement. That is available to you at any scale.<\/p>\n\n\n\n Same day, same religion, completely different answer. Meta’s second-quarter release<\/a> shows revenue of $60.8 billion, up 28% and ahead of estimates, with net income of $15.8 billion and diluted EPS of $6.18 against a roughly $7.13 consensus. Capital expenditures including finance leases were $31.1 billion for the quarter. Free cash flow was $784 million. A year earlier it was $8.55 billion, which makes this a 91% collapse, and total expenses grew 55% partly on $2.4 billion of legal charges and severance. Meta raised the floor of its full-year capex range to $130 billion and guided full-year expenses to $165 billion to $169 billion. The stock fell about 10% after hours, and CNBC’s read tied the drop to guidance and the cash flow line<\/a> rather than the revenue beat.<\/p>\n\n\n\n Verdict: matters, and it is the cleanest AI capex comparison you will get all year.<\/strong> Two of the largest spenders reported within hours of each other. One showed accelerating revenue from the buildout. The other showed a 91% cash-flow collapse with the payoff still described in future tense. When someone tells you AI infrastructure spending is either obviously justified or obviously a bubble, this pair is the answer: it depends entirely on whether you have somewhere to sell the capacity.<\/p>\n\n\n\nMicrosoft Grew Azure Faster by Working the Fleet It Already Owns<\/h2>\n\n\n\n
Meta Spent $31 Billion in One Quarter and Kept $784 Million<\/h2>\n\n\n\n
Zuckerberg’s Enterprise Pitch Is the Ad Model Wearing an Agent Costume<\/h2>\n\n\n\n