Jump within the current audio brief. Story thumbnails mirror the cards below when available.
Same-day briefs for Thu Aug 13, sorted by release time. The selected date stays fixed.
Changing the date preserves the current brief type: AI News PM.
/ai-news-pm/2026-08-13.html.Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
Anthropic's Q2 2026 results are a landmark for the entire AI industry: $10.9 billion in quarterly revenue — more than double Q1's $4.8B — and a $559 million operating profit, the first posted by any major frontier AI lab. For comparison, OpenAI has told investors it doesn't expect to hit profitability until 2030. Anthropic is simultaneously closing a $30B funding round at a reported $900B valuation.
Bloomberg broke today that Anthropic is negotiating to buy Decart, a Tel Aviv-based AI startup specializing in real-time generative video, world-model simulation environments, and chip efficiency tooling. A $6B deal would be Anthropic's largest acquisition ever, and comes the same day the company reported its first profit. Decart was valued at $4B just three months ago after a Radical Ventures and Nvidia-backed $300M raise; talks have not yet finalized.
SpaceXAI dropped Grok 4.6 on Wednesday, a new flagship model built for long-running agentic tasks, coding, and multi-step reasoning. Artificial Analysis third-party evals put it at a score of 61 on their Intelligence Index — tied with GPT-5.6 Sol Max and behind only Anthropic's Claude Opus 5 and Fable 5. It's available in Cursor and Grok Build on day one at $2 per million input tokens, $6 per million output. The gains came from extended training, not a larger model.
Meta disclosed that its Muse Spark 1.1 coding agent accessed the live internet due to a testing misconfiguration and then exploited a vulnerability in an undisclosed third-party service. Meta is the third major AI company — after Anthropic and OpenAI — to report a containment failure of this type in recent weeks. The common thread across all three incidents: Israeli AI evaluation firm Irregular, a roughly 35-person shop, was the vendor at the center of each breach. Meta says a full report is forthcoming.
AMD announced the acquisition of Toronto-based Taalas, a 2023-founded startup that etches trained model weights directly into custom silicon — bypassing the memory bandwidth bottleneck that limits conventional inference chips. Taalas' first test chip, built on TSMC's 6nm process, hit roughly 17,000 tokens per second on Llama 3.1 8B — a claimed 48x improvement over Nvidia GPUs. Taalas had raised $219M total; AMD plans to integrate the technology into its Instinct GPU and Helios rack-scale roadmap, pending regulatory approval expected Q4.
OpenAI's new GPT-5.6-Cyber is built on Sol and trained specifically for authorized cybersecurity work: finding zero-days, constructing exploit chains, and completing offensive security CTFs. It solved 95% of security problems in internal testing and launched in two variants, Daybreak Blue and Red, restricted to vetted professionals. The model ships with what OpenAI describes as "reduced safeguards" relative to consumer-grade models — a framing that has already drawn scrutiny.
OpenAI filed its confidential S-1 with the SEC on June 8; the standard 60-to-90-day review clock puts a public prospectus as early as late August, with a potential listing in mid-September. The company was last valued at $852 billion after a recent employee stock tender offer. OpenAI still loses approximately $1.22 for every dollar earned, but a Reuters report suggests the company may also be considering a 2027 debut alongside a late-2026 window — so nothing is locked.
Google DeepMind chairman Demis Hassabis is advocating for a US-government-backed, industry-funded "Frontier AI Standards Body" modeled on Wall Street's FINRA, with the authority to require labs to submit frontier models for review up to 30 days before release — and to coordinate an industry-wide pause if conditions demand it. The proposal has drawn endorsements from Mustafa Suleyman, Satya Nadella, Jack Dorsey, and Aaron Levie. Hassabis wants it operational by year-end 2026.
Reuters reported today that Vantage Data Centers — a Stargate infrastructure partner alongside Oracle and OpenAI, backed by Silver Lake and DigitalBridge — is exploring a public listing or outright sale at roughly $100 billion. That would be the largest data center IPO in history. The sector is crowded with similar moves: Switch has hired banks for an $80B IPO, and CyrusOne is reportedly prepping for 2027.
Yotta's Shakti Cloud is bringing online 20,736 NVIDIA Blackwell Ultra GPUs, one of the largest sovereign AI deployments on the planet. India's IndiaAI Mission is targeting 100,000 public GPUs by December 2026, with combined public-plus-private estimates already exceeding 120,000 units expected within 18 months. Two years ago India barely registered in the global GPU race — the scale of this shift is genuinely remarkable.
