LastWeekIn.Tech

Today's AI news is dominated by frontier-lab economics and infrastructure spending — Anthropic's IPO filing and Microsoft's headcount-for-datacenters trade reveal the scale of the compute buildout — alongside a notable policy alignment on web scraping and fresh evidence on AI security risks.

1. 1

   Number 1. Industry & Economics buttondown.com

   Anthropic files for an IPO, revealing frontier-lab economics

   Anthropic's IPO filing (S-1) discloses $16.29 billion in first-half 2026 revenue against $2.38 billion in cost of revenue, an 85% gross margin, with almost all costs tied to compute and infrastructure. The filing reveals roughly $750 billion in computing commitments, including up to $120.5 billion to SpaceX's AI subsidiary through 2029, and describes a history of operating in a compute-constrained environment.

   Why it matters This is the most detailed public look yet at how a frontier AI lab's unit economics actually work, showing massive revenue growth paired with even more massive forward compute commitments that dwarf current revenue. It gives investors, regulators, and competitors a template for evaluating whether AI labs are genuinely profitable businesses or compute-financing vehicles, and will likely set the benchmark against which OpenAI's and others' disclosures are judged.

   What to watch Watch for how public markets price Anthropic's IPO once it prices, and whether rivals like OpenAI face pressure to disclose similarly granular compute-cost breakdowns.

2. 2

   Number 2. Policy & Regulation iheart.com

   OpenAI reportedly raises $11 billion at a $1.25 trillion valuation

   Google, Microsoft, OpenAI, and Anthropic are backing a unified robots.txt-style signal that lets publishers block AI scraping across all four platforms simultaneously. The move comes amid mounting legal pressure over unauthorized training-data collection.

   Why it matters A shared opt-out standard across the four largest AI developers would be the first coordinated industry response to publisher and rights-holder complaints about scraping, potentially reducing the patchwork of lawsuits and site-specific blocking rules that have proliferated over the past two years. It signals the industry anticipates further regulatory and legal scrutiny of training-data sourcing and wants to self-regulate before mandates arrive.

   What to watch Watch whether publishers and news organizations adopt the standard widely, and whether smaller AI labs or Chinese model developers agree to honor it.

3. 3

   Number 3. Infrastructure & Chips iheart.com

   Microsoft reportedly offers voluntary severance to fund AI infrastructure

   Microsoft is offering voluntary severance packages to roughly a quarter of its workforce — about 60,000 employees — while protecting AI-critical teams from cuts. The reallocated capital is intended to help fund a $170 billion AI infrastructure buildout.

   Why it matters This is one of the clearest signals yet that a major tech company is restructuring its entire workforce strategy around AI infrastructure spending, treating headcount reduction as a direct funding mechanism for data centers rather than a response to weak demand elsewhere. It underscores how capital-intensive the AI buildout has become and raises questions about whether other hyperscalers will follow with similar trade-offs between labor and compute investment.

   What to watch Watch for Q4 earnings commentary on whether severance savings materially offset data center capex, and whether other hyperscalers announce comparable workforce-for-compute tradeoffs.

4. 4

   Number 4. Infrastructure & Chips aiweekly.co

   Moonshot restricts new API-user access to its Kimi K2 model series

   SemiAnalysis benchmark data published October 31 shows AMD's MI355X serves Moonshot's 2.8-trillion-parameter Kimi K3 model at $0.077 per million tokens versus $0.179 on Nvidia's B200 at 100 tokens/sec per user — about 2.3x cheaper. The gap narrows to $0.146 vs $0.275 at higher throughput, with the advantage attributed to the MI355X's larger HBM capacity and kernel-level tuning.

   Why it matters Independent, third-party benchmarks showing AMD meaningfully undercutting Nvidia on inference cost-per-token for a large open model are rare and significant, given Nvidia's near-total dominance of AI inference infrastructure. If these results hold up at scale, they could accelerate multi-vendor chip strategies among hyperscalers and give large open-weight model providers like Moonshot leverage to negotiate better hosting economics.

   What to watch Watch for hyperscaler adoption announcements of MI355X at scale and whether Nvidia responds with pricing or performance updates for B200 successors.

5. 5

   Number 5. Security & Safety Hacker News

   Brazilian Court Sanctions Lawyers for Prompt Injection in Court Filing

   A Brazilian labor court censured and fined two lawyers for intentionally embedding a hidden prompt-injection command in a court filing, aiming to manipulate the AI system courts use to review filings. This may be the first judicial ruling globally to formally address intentional prompt-injection manipulation of a court's AI system.

   Why it matters As courts, governments, and businesses increasingly deploy AI to review documents and filings, this ruling establishes real-world legal precedent that prompt injection against institutional AI systems is sanctionable misconduct, not just a theoretical security concern. It signals that legal systems are beginning to treat AI manipulation attempts with the same seriousness as other forms of court interference or fraud.

   What to watch Watch for similar cases in other jurisdictions and whether bar associations or court systems issue formal guidance on AI-filing manipulation as a professional-conduct violation.