Thursday tilted the AI landscape in the industry's favor on three fronts at once: money, courts, and rules. Broadcom posted the kind of quarter that only the AI build-out can produce, Google escaped its second threatened breakup of the year, and both the G20 and the U.S. Justice Department signaled that the regulatory brakes are coming off.
Broadcom's AI chip business triples to $16.7 billion
- AI semiconductor revenue hit $16.7 billion in fiscal Q3, up 221% year-over-year, driven by custom accelerators (XPUs) and networking chips for hyperscalers; total revenue reached $29.6 billion.
- The company guided to roughly $34.8 billion in AI chip revenue for Q4 alone β more than double this quarter β as demand outstrips its own multi-year roadmap.
Why it mattersWhy it matters: custom silicon is now the credible alternative to Nvidia, and the buildout budgets behind these numbers will shape cloud prices everywhere β including what Latin American startups pay for compute.
Google keeps AdX but must rewrite its auction rules
- Judge Leonie Brinkema rejected the DOJ's demand that Google sell its AdX ad exchange, citing the lack of an obvious buyer; the 2025 liability ruling had found Google held 91% of the publisher ad server market.
- Instead, Google faces six years of monitored behavioral remedies: sharing real-time AdX bid data with rival ad servers, ending unified pricing rules, and a ban on 'first look' and 'last look' advantages.
Why it mattersWhy it matters: publishers β including Latin American media that depend on Google's ad stack β gain real control over their auctions, while Google dodges its second breakup attempt this year.
Meta says Muse Spark 1.3 has caught up with the frontier
- Meta released Muse Spark 1.3, claiming it matches Anthropic's and OpenAI's frontier models and beats them at coding while using about 25% fewer tokens per task; its stock rose roughly 4% on the news.
- Meta plans to publish open weights for version 1.2 but hasn't decided for 1.3 β a choice with direct consequences for EU AI Act compliance and for developers who build on open models.
Why it mattersWhy it matters: if the claims hold, the frontier is now a three-way race β and cheaper tokens for agents is exactly the battleground where the next year of AI economics gets decided.
The G20 endorses Washington's light-touch AI rulebook
- All 20 members β including China and Russia β backed the U.S.-drafted 'Carolina Principles' at the Chapel Hill summit: sector-by-sector regulation, no new AI agencies, and close collaboration with private companies.
- The accord is non-binding, with formal adoption expected in December at the G20 gathering in Doral, Florida; Jensen Huang, Sam Altman, Mark Zuckerberg and Elon Musk attended the summit.
Why it mattersWhy it matters: the global regulatory wind just shifted toward deregulation β a signal Latin American legislators drafting AI laws, from Brazil to Chile, will not ignore.
The DOJ sides with OpenAI in the New York Times lawsuit
- The Justice Department filed a statement of interest backing OpenAI and Microsoft, arguing that training AI on copyrighted text is transformative fair use β the U.S. government's first intervention in an AI training lawsuit.
- The brief warns that restricting AI training would harm innovation, economic growth and national security, and suggests creators seeking compensation should turn to Congress, not the courts.
Why it mattersWhy it matters: dozens of copyright suits against AI labs now face a government thumb on the scale β if fair use prevails, the training-data question moves from courtrooms to Congress.
The big picture
In a single day, the scales tipped toward the AI industry on money, courts and rules at once. Broadcom proved the buildout is still accelerating, Google and OpenAI got favorable signals from the bench, and the G20 blessed a hands-off approach. The build-out now has more capital and fewer brakes than at any point this year.