SpaceX officially unveiled its orbital data center project for AI, named Starmind, on July 7, 2026. The first satellites, designated AI1, will feature 70-meter solar panels capable of generating an average of 120 kW (up to 150 kW peak) and a 110 m² radiator with cooling fluid to dissipate heat from the GPUs into the vacuum of space.

The economic rationale behind these space data centers includes free and unlimited solar energy in orbit, eliminating the need for land, cooling water, or grid connections. Each satellite will house up to twelve independent compute modules, utilizing AI processors from various chip manufacturers, including Nvidia. These satellites will operate in a heliosynchronous polar orbit (SSO) for constant solar illumination and will be interconnected via laser links, similar to the Starlink constellation.

In parallel, SpaceX has requested FCC authorization to operate up to 100,000 third-generation Starlink satellites, potentially to support Starmind's data transmission needs. The economic viability of the project heavily depends on SpaceX's ability to drastically reduce launch costs with Starship, with manufacturing planned at the new Gigasat factory in Bastrop, Texas. Concerns have also been raised regarding the impact of such a large number of satellites on the night sky and astronomy.

👉 Read the full article on Gizmodo en Español

OpenAI has publicly launched its new family of AI models, GPT-5.6, comprising three variants: Sol (the flagship model), Terra (a balanced option), and Luna (the fastest and most economical alternative). GPT-5.6 Sol is described as the most capable generalist model developed by the company to date, designed for complex problem-solving, tool utilization, internet navigation, programming, and sustained reasoning processes.

Independent tests position GPT-5.6 Sol among the most capable systems on the market, notably leading the Coding Agent Index with a score of 80 for autonomous terminal use and error resolution. It also became the first frontier model to win a public game in ARC-AGI-3, demonstrating a formidable ability to interpret unknown interactive scenarios, formulate hypotheses, and adapt strategies.

However, the article emphasizes that for most daily tasks, a powerful model like Sol is often unnecessary, and faster, cheaper options like Terra or Luna suffice. Sol costs $5 per million input tokens and $30 per million output tokens, significantly more than Luna's $1 and $6 respectively. This suggests that automatically choosing the most powerful model may not be the most economically intelligent decision, highlighting the importance of selecting a model with sufficient intelligence for the specific task.

👉 Read the full article on Gizmodo en Español

This article focuses on scientists' research into combining AI and quantum computing to generate novel peptides. This integrated approach holds the potential to revolutionize fields such as drug discovery, materials science, and biotechnology.

AI, particularly generative AI models, is employed to predict, design, and optimize peptide sequences with desired properties. Concurrently, quantum computing accelerates simulations of molecular interactions, folding, and energy landscapes, which are computationally intensive for classical computers, leading to more accurate predictions and faster discovery.

The advancement of this technology is expected to contribute to the development of new therapeutics and antibiotics, the creation of innovative biomaterials, and the design of enzymes and biosensors. However, challenges remain in effectively integrating AI and quantum computing, experimentally validating the generated peptides, and scaling quantum solutions.

👉 Read the full article on Wired