Google is working on an ambitious project called Starcatcher, aiming to place massive AI data centers in space. These data centers would be powered by satellites equipped with solar panels positioned in optimal solar orbits, where solar energy can be up to eight times more productive than on Earth. This approach could solve the current energy bottleneck faced by terrestrial data centers, especially in the U.S., where some GPUs remain unused due to power constraints. The project involves closely clustered satellites communicating via high-bandwidth intersatellite links, overcoming challenges like radiation exposure and the high cost of launching materials into orbit, which is expected to decrease significantly by the 2030s.
OpenAI is aggressively expanding its compute infrastructure, having signed massive deals worth hundreds of billions with companies like Nvidia, AMD, Amazon, Intel, TSMC, Microsoft, and Oracle. Recently, OpenAI lost Microsoft’s right of first refusal for compute resources and quickly partnered with Amazon AWS to secure additional capacity. This shift highlights the intense competition and high demand for compute power in the AI industry, as companies race to develop and deploy increasingly complex models. OpenAI is also reportedly considering a trillion-dollar IPO, reflecting its rapid growth and market influence.
Apple is reportedly planning to enhance Siri by integrating a custom version of Google’s Gemini AI model, paying Google around $1 billion annually for the license. Unlike previous setups where Siri relied on third-party models like OpenAI’s with a clunky user experience, Apple will run Gemini on its private cloud infrastructure, allowing better control and optimization. This new Siri is expected to launch next spring and represents a significant step forward in Apple’s AI capabilities, leveraging a longstanding partnership with Google while maintaining control over the user experience.
Microsoft has introduced a new in-house image generation model called MAI Image One, which ranks among the top 10 text-to-image models on LM Arena. The model focuses on delivering high-quality, photorealistic images with improved creativity and less generic output, addressing a common criticism of AI-generated content. Users can try it out on Bing’s image creation platform. Additionally, a new video generation model, VU Q2, has emerged as a strong contender on the text-to-video leaderboard, offering up to 8 seconds of 1080p video generation with support for multiple reference images to enhance control and quality.
In other news, Amazon has issued a legal threat against Perplexity’s AI-first browser, Comet, aiming to prevent users from using AI assistants to shop on Amazon’s platform. This move reflects Amazon’s concern over AI agents bypassing traditional advertising models by directly purchasing products on behalf of users. Meanwhile, OpenAI’s text-to-video platform Sora has introduced a new feature allowing users to upload custom characters or pets to cameo in videos, enhancing personalization and creativity. These developments illustrate the evolving landscape of AI tools and the ongoing tensions between innovation and existing business models.
