Strategy
177 analyses
Review of the MCP published on July 28 removes persistent session and initialization handshake. Servers now run on Lambda, Cloud Function, and AgentCore, scaling to zero.
In a post published on Monday, the CEO of Anthropic separates the defense of open models from the specific fear of Kimi K3 and shifts the debate towards chips, distillation, and mandatory safety testing.
Nvidia is discussing securing $250 billion in funding for OpenAI to lease a 10 GW campus in Piketon, as revealed by the Wall Street Journal, in an arrangement that exposes the circular financing of AI.
Moonshot AI released the Kimi K3 weights on Hugging Face today at midnight UTC. With 2.8 trillion parameters and a modified MIT license, it is the first open model in the 3 trillion range, forcing U.S. hyperscalers to rethink their proprietary strategies.
On the morning of July 24, AWS experienced a 64-minute outage in the Oregon region, the third failure in three months, affecting Apple Pay, DoorDash, Reddit, PlayStation Network and other services.
Launched on July 24, Claude Opus 5 delivers performance close to Fable 5 at $5 per million input tokens, with effort control at five levels and a context window of 1 million tokens.
A network configuration change degraded over 20 Azure services for approximately four hours on July 23, exposing the fragility of multi-region single-cloud architectures amid the hyperscaler race.
The Presence platform enters limited availability with BBVA, SoftBank, and IAG as its first clients, while Project Camellia adds 3.2 GW to the race for data center capacity in the United States.
Cuts impact post-training teams and model customization within the Nova line while the company maintains a $200 billion capex guide and redirects funds to engineers embedded with corporate clients.
A deal for 2 gigawatts of Instinct MI450 GPUs and equity investment tied to milestones positions AMD as the fourth supplier for Anthropic, alongside AWS Trainium, Google TPU, and Nvidia.
Microsoft and Mistral finalized a multibillion-dollar partnership expansion on Tuesday, with Medium 3.5 and OCR 4 arriving at Foundry, Copilot Studio, and Azure Local hardware capable of running offline.
Google launched Gemini 3.6 Flash at $1.50 per million input tokens and $7.50 for output, generating 17% fewer tokens. On the same day, DeepMind confirmed the start of the largest pre-training in history.