# AI News Digest - 2026-05-07

1. Anthropic took over the full computing capacity of SpaceX's Colossus 1 data center, securing more than 300 megawatts and over 220,000 NVIDIA GPUs expected to come online within a month, and it doubled rate limits for Claude Code while raising API limits for Opus models; the company also committed roughly $200 billion to Google Cloud over five years.

2. OpenAI swapped ChatGPT's default model for GPT-5.5 Instant, which produced 52.5 percent fewer hallucinated claims on high-risk topics in internal testing and introduced "memory sources" with phased personalization for Plus and Pro web users.

3. OpenAI built an open-source networking protocol called MRC with AMD, Broadcom, Intel, Microsoft, and NVIDIA to send data across hundreds of paths simultaneously between GPUs, reducing required switch layers to two for over 100,000 GPUs and cutting power use and costs; the protocol was running on OpenAI's Stargate supercomputer.

4. US Department of Commerce obtained pre-release access to AI models from Anthropic, OpenAI, Google DeepMind, Microsoft, and xAI for classified national security testing, with the companies providing models with reduced safety guardrails for testing in secure environments.

5. Google released multi-token prediction drafters for its Gemma 4 model family that used a small auxiliary model to suggest several tokens at once while the main model checked them in a single pass, speeding up text generation by up to three times.

# References

1. [https://the-decoder.com/anthropic-taps-spacexs-colossus-1-data-center-for-220000-gpus-to-power-claude/](https://the-decoder.com/anthropic-taps-spacexs-colossus-1-data-center-for-220000-gpus-to-power-claude/) [https://the-decoder.com/anthropic-commits-200-billion-to-google-cloud-over-five-years/](https://the-decoder.com/anthropic-commits-200-billion-to-google-cloud-over-five-years/)

[Anthropic commits $200 billion to Google Cloud over five years](https://the-decoder.com/anthropic-commits-200-billion-to-google-cloud-over-five-years/)

[Anthropic taps SpaceX's Colossus-1 data center for 220,000 GPUs to power Claude](https://the-decoder.com/anthropic-taps-spacexs-colossus-1-data-center-for-220000-gpus-to-power-claude/)

1. [https://the-decoder.com/chatgpt-update-rolls-out-gpt-5-5-instant-with-fewer-hallucinations-and-more-personalized-answers/](https://the-decoder.com/chatgpt-update-rolls-out-gpt-5-5-instant-with-fewer-hallucinations-and-more-personalized-answers/)

[ChatGPT update rolls out GPT-5.5 Instant with fewer hallucinations and more personalized answers](https://the-decoder.com/chatgpt-update-rolls-out-gpt-5-5-instant-with-fewer-hallucinations-and-more-personalized-answers/)

1. [https://the-decoder.com/openai-built-a-networking-protocol-with-amd-broadcom-intel-microsoft-and-nvidia-to-fix-ai-supercomputer-bottlenecks/](https://the-decoder.com/openai-built-a-networking-protocol-with-amd-broadcom-intel-microsoft-and-nvidia-to-fix-ai-supercomputer-bottlenecks/)

[OpenAI built a networking protocol with AMD, Broadcom, Intel, Microsoft, and NVIDIA to fix AI supercomputer bottlenecks](https://the-decoder.com/openai-built-a-networking-protocol-with-amd-broadcom-intel-microsoft-and-nvidia-to-fix-ai-supercomputer-bottlenecks/)

1. [https://the-decoder.com/us-government-now-has-pre-release-access-to-ai-models-from-five-major-labs-for-national-security-testing/](https://the-decoder.com/us-government-now-has-pre-release-access-to-ai-models-from-five-major-labs-for-national-security-testing/)

[US government now has pre-release access to AI models from five major labs for national security testing](https://the-decoder.com/us-government-now-has-pre-release-access-to-ai-models-from-five-major-labs-for-national-security-testing/)

1. [https://the-decoder.com/google-speeds-up-gemma-4-threefold-with-multi-token-prediction/](https://the-decoder.com/google-speeds-up-gemma-4-threefold-with-multi-token-prediction/)

[Google speeds up Gemma 4 threefold with multi-token prediction](https://the-decoder.com/google-speeds-up-gemma-4-threefold-with-multi-token-prediction/)

For the site tree, see the [root Markdown](https://ixtj.dev/.md).