The Vikshy Brief · Saturday 26 September

The day in AI, in five minutes.

Anthropic reportedly signed a seven-year, $11.6 billion cloud deal with Akamai, while OpenAI reported a Codex outage caused by an internal backend key error. DeepSeek moved from #7 to #6 on the Vikshy Score, and Z.ai moved from #6 to #7.

4 stories · published 19:12 UTC · · The lede and the "why it matters" lines are written by GPT-6 Luna from the facts below only; everything else is read from the sources.

What mattered

  1. 1
    AnthropicMajor money news25 Sep

    Anthropic signs $11.6 billion cloud deal with Akamai

    Why it matters: The reported deal adds to Anthropic's reported $517 billion in compute deals over 11 months and includes a warrant for up to 5 percent of Akamai's shares.

    Anthropic has reportedly signed a seven-year, $11.6 billion cloud deal with Akamai Technologies and will receive a warrant for up to 5 percent of Akamai's shares. Its compute deals have reportedly totaled $517 billion in 11 months.

  2. 2
    OpenAIMajor outage25 Sep

    OpenAI Codex outage caused by backend key error

    Why it matters: The internal 401 backend key error disrupted Codex, and OpenAI said it had identified the issue and was working toward mitigation.

    OpenAI reported a Codex outage caused by an internal 401 backend key error. The company said it identified the issue and is working toward mitigation.

  3. 3
    AmazonResearch25 Sep

    Amazon details MoE reinforcement learning scaling on EKS with EFA and DeepEP

    Why it matters: Amazon reports that its EKS, EFA and S3 architecture increased aggregate reinforcement learning rollout throughput by 40 percent for RLHF and GRPO training.

    Amazon published an architecture combining Amazon EKS, Elastic Fabric Adapter and Amazon S3 for large-scale MoE reinforcement learning. It reports a 40% increase in aggregate reinforcement learning rollout throughput for RLHF and GRPO training.

  4. 4
    AmazonNew product or feature25 Sep

    Amazon SageMaker AI adds Qwen3-TTS real-time speech deployment

    Why it matters: SageMaker JumpStart now offers a managed real-time endpoint for Qwen3-TTS, including voice cloning from a short clip and cross-lingual speaker identity preservation.

    Amazon SageMaker JumpStart now supports deploying the publicly available Qwen3-TTS-12Hz-1.7B-Base text-to-speech model to a fully managed real-time endpoint. The deployment supports voice cloning from a short reference clip, with cross-lingual cloning preserving speaker identity across languages.

The Vikshy Score

The close of 26 Sep, 06:00 UTC, against the previous close. The full board.