Anthropic IPO Filing Discloses $895.4 Billion Compute Commitment · 1103-793
In its IPO filing, Anthropic disclosed for the first time compute contract commitments of up to $895.4 billion, while major investors hope to set its valuation in the $1.5 trillion to $1.6 trillion range [0][14]. Meanwhile, OpenAI is delivering AI governance memos to 2028 election candidates, while Altman claims the next-generation GPT has proven a Millennium Prize Problem [1][11]. On the technical side, DeepSeek V4.1 Flash scored 78.1% on ARC-AGI-2 at a per-task cost of only...
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Core Insights
**Anthropic** disclosed for the first time in its IPO filing compute contract commitments of up to **$895.4 billion**, while major investors hope to set its valuation in the **$1.5 trillion to $1.6 trillion** range [0][14]. Meanwhile, **OpenAI** is delivering AI governance memos to 2028 election candidates, while **Altman** claims the next-generation GPT has proven a Millennium Prize Problem [1][11]. On the technical side, **DeepSeek V4.1 Flash** scored **78.1%** on ARC-AGI-2 at a per-task cost of only **$0.16**, and **agent memory architectures** and the open-sourcing of **Super-agent** became this week's research hotspots [17][19].
## 🚀 Key Developments
- **Anthropic IPO valuation expectations lowered** [0]: Major investors hope to set the valuation at $1.5 trillion to $1.6 trillion, below the previously discussed figure of nearly $2 trillion.
- **Anthropic discloses $895.4 billion in compute contract commitments** [14]: Of this, $447.5 billion must be paid by 2031, and only $213.8 billion is cancellable or conditional.
- **OpenAI distributes AI memo to 2028 election candidates** [1]: Lehan says AI will be a mandatory question in the presidential campaign, and candidates need to present a complete governance vision.
- **Altman says next-generation GPT has proven a Millennium Prize Problem** [11]: He says 10,000 copies running continuously for about 4 days completed the proof, with no hierarchy among the copies.
- **DeepSeek V4.1 Flash tops ARC-AGI cost-effectiveness** [17]: ARC-AGI-2 score of 78.1% at $0.16 per task; ARC-AGI-1 score of 94.5% at $0.06 per task.
- **DeepSeek open-sources Super-agent architecture** [19]: A Router-Worker two-tier design that enables AI to assign work autonomously, proactively seek work, and continuously follow up.
- **Trajectory closes $45 million Series A** [2]: Led by Sequoia with participation from Nvidia, Bessemer, and Bain Capital, at a $300 million valuation, focused on enabling non-technical experts to train models.
- **Goldman Sachs restricts traders' use of AI coding tools** [23]: Starting in October, permissions for Claude Code, GitHub, and others will be narrowed to engineers; traders and bankers will no longer have access.
- **OpenAI deploys activation classifiers for safety monitoring** [24]: Since GPT-5.6, real-time per-token scoring can block violating responses mid-generation and hand them to a reasoning model for review.
- **Cohere paper overturns old assumptions about speculative decoding** [15]: On fine-grained MoE, gains remain significant even at batch size 1.
## 🔗 Sources
[0] WSJ: Major investors hope to set Anthropic's IPO valuation at $1.5 trillion to $1.6 trillion — https://aihot.news/items/p61czu93us1j0lli5em6rjkte
[1] OpenAI distributes memo to potential 2028 US election candidates, saying AI will be a mandatory question — https://aihot.news/items/bt4v2dd852q3dl0xtndpbfc3p
[2] Trajectory raises $45 million to let non-technical experts train AI models — https://aihot.news/items/gmapr5adh7u2fbqy69sesf6se
[11] Altman says
Anthropic disclosed for the first time in its IPO filing compute contract commitments of up to $895.4 billion, while major investors hope to set its valuation in the $1.5 trillion to $1.6 trillion range [0][14]. Meanwhile, OpenAI is delivering AI governance memos to 2028 election candidates, while Altman claims the next-generation GPT has proven a Millennium Prize Problem [1][11]. On the technical side, DeepSeek V4.1 Flash scored 78.1% on ARC-AGI-2 at a per-task cost of only $0.16, and agent memory architectures and the open-sourcing of Super-agent became this week's research hotspots [17][19].
🚀 Key Developments
- Anthropic IPO valuation expectations lowered [0]: Major investors hope to set the valuation at $1.5 trillion to $1.6 trillion, below the previously discussed figure of nearly $2 trillion.
- Anthropic discloses $895.4 billion in compute contract commitments [14]: Of this, $447.5 billion must be paid by 2031, and only $213.8 billion is cancellable or conditional.
- OpenAI distributes AI memo to 2028 election candidates [1]: Lehan says AI will be a mandatory question in the presidential campaign, and candidates need to present a complete governance vision.
- Altman says next-generation GPT has proven a Millennium Prize Problem [11]: He says 10,000 copies running continuously for about 4 days completed the proof, with no hierarchy among the copies.
- DeepSeek V4.1 Flash tops ARC-AGI cost-effectiveness [17]: ARC-AGI-2 score of 78.1% at $0.16 per task; ARC-AGI-1 score of 94.5% at $0.06 per task.
- DeepSeek open-sources Super-agent architecture [19]: A Router-Worker two-tier design that enables AI to assign work autonomously, proactively seek work, and continuously follow up.
- Trajectory closes $45 million Series A [2]: Led by Sequoia with participation from Nvidia, Bessemer, and Bain Capital, at a $300 million valuation, focused on enabling non-technical experts to train models.
- Goldman Sachs restricts traders' use of AI coding tools [23]: Starting in October, permissions for Claude Code, GitHub, and others will be narrowed to engineers; traders and bankers will no longer have access.
- OpenAI deploys activation classifiers for safety monitoring [24]: Since GPT-5.6, real-time per-token scoring can block violating responses mid-generation and hand them to a reasoning model for review.
- Cohere paper overturns old assumptions about speculative decoding [15]: On fine-grained MoE, gains remain significant even at batch size 1.
🔗 Sources
[0] WSJ: Major investors hope to set Anthropic's IPO valuation at $1.5 trillion to $1.6 trillion — https://aihot.news/items/p61czu93us1j0lli5em6rjkte [1] OpenAI distributes memo to potential 2028 US election candidates, saying AI will be a mandatory question — https://aihot.news/items/bt4v2dd852q3dl0xtndpbfc3p [2] Trajectory raises $45 million to let non-technical experts train AI models — https://aihot.news/items/gmapr5adh7u2fbqy69sesf6se [11] Altman says