noticias ia para empresas meta ads anthropic

Platform contracts are back in the news: Meta has frozen, with no new date, the terms that would have let it crawl any advertiser's website and made that advertiser answer for whatever its AI changes in the ads. The sector's money is flowing into power and new hardware: Anthropic is paying $3.4 billion for gas generators, and Microsoft is walking away from Oracle servers running older chips. And Thinking Machines, which is negotiating a $5 billion round, has unveiled an open model that's cheap per task but burns four times as many tokens as its predecessor. Catch up on Sunday's radar.

Meta freezes the contract that let it crawl your site and blame you for what its AI changes in your ads

On Thursday, November 5, Meta paused, with no new date, an update to four contracts every advertiser would have accepted simply by continuing to use its tools: the Commercial Terms, Product Catalog Terms, Self-Serve Ad Terms and Ad Creative Generative AI Terms. They were due to take effect on November 20; according to a notice in Meta's Business Help Center, they won't. The current terms stay in force, which for the Commercial Terms means the March 2024 version. The frozen text authorized Meta to send its bots to the websites linked to any business using its products, and it added a section making the advertiser solely responsible for AI changes to its ads, including features Meta switches on by itself. That's not a niche group: nearly 10 million advertisers used at least one of Meta's generative AI creative tools last quarter. The pause follows an injunction from an Oslo court, another from São Paulo carrying a daily fine of one million reais, and an investigation by the UK competition regulator, which is still open. Meta calls it a pause, not a withdrawal, and the notice says nothing about its crawlers or how it treats robots.txt. What it means for your business: nothing changes on November 20, but the clause could come back. Use the breathing room to check which automatic Advantage+ features you have switched on (copy, translations, generated images) and turn off anything nobody reviews before it goes live. Decide in your robots.txt which AI bots can get into your site. And save a copy of today's terms so you can compare when Meta sends the new ones. An AI that publishes in your name without review is the UNATTENDED rung with no safety net: only climb there once you know who answers for each mistake. Source

Anthropic pays $3.4 billion for generators and Microsoft drops old Oracle servers: AI's limit is now power

Anthropic has agreed to pay $3.4 billion for the rights to orders for Caterpillar gas generators with 1.7 gigawatts of capacity, the Financial Times reported on Sunday, calling it the largest known purchase of power-generation equipment by an AI company. The orders were placed about two years ago by OnePWR Solutions, a San Antonio company; Anthropic takes over those orders along with OnePWR's engineering and installation capacity, but not the company itself. The deal was signed in late September, there's no word on where the generators will go or when they'll run, and the reason is lead time: a new generator or turbine now takes years to arrive. On Friday, Bloomberg reported that Microsoft is letting a 2023 contract to rent AI capacity from Oracle for Bing run out without renewing it, because its older-generation Nvidia chips no longer meet Microsoft's specs, as part of a review of more than $80 billion in commitments to outside cloud providers. Oracle's response: chip capacity that came up for renewal in its latest fiscal quarter was renewed or resold at a 20% premium on long-term contracts. What it means for your business: the giants are locking in power years ahead and discarding older hardware, so good capacity is likely to get scarce before it gets cheap. Our take: ask your provider which region the model you use runs in and what availability it commits to, and keep a tested second provider for the process that brings in the most revenue. An INTEGRATED process that hangs on a single model with no plan B isn't ready to go UNATTENDED. Source and Source

Thinking Machines eyes $5 billion and launches an open model: cheap per task, heavy on tokens

The company run by Mira Murati, OpenAI's former chief technology officer, is negotiating a $5 billion round led by Accel at a $40 billion pre-money valuation, according to Business Insider; it hasn't closed and hasn't been confirmed. In July 2025 it raised $2 billion at $12 billion, and early this year an attempt to raise at $50 billion to $60 billion fell apart. At the same time, it has unveiled Inkling 2 Small: 594 billion parameters, about 17 billion of them active per query, available on its Tinker cloud, with open weights promised for this week. Artificial Analysis ranks it as the best open model built outside China, provided the weights actually ship. The cost comes with fine print: it produces about 150,000 output tokens per index task, four times its predecessor, and each task costs $1.18, almost six times as much as before, though far below Gemini 3.8 Flash at $7.77 or Claude Sonnet 5.5 at $10.02. And its lower hallucination rate comes mostly from answering less: it attempts 53% of questions, down from 77%. What it means for your business: compare models on cost per finished task, not price per token, as we saw with spend per employee. If the weights come out, it's a way to run the model with a provider you choose, but at that size it won't fit on an office server. Test it on ten real tasks and count how many it leaves unanswered. Source and Source

What to watch this week

Whether Thinking Machines releases the Inkling 2 Small weights and closes its round, whether Anthropic kicks off its IPO roadshow, and whether Meta spells out which parts of its terms it will rewrite. Also today, Monday the 9th, Copilot Chat caps reach every organization.