Meta Follows SpaceX's Playbook, Plans to Sell Surplus AI Compute as a Cloud Business

Meta is reportedly building a cloud business to sell the surplus capacity of its massive AI compute resources to external customers. It's the same approach SpaceX took when it offered infrastructure built up through rocket development to outside clients.

For Meta, which plans to invest up to $145 billion (roughly 22 trillion yen) in AI during 2026, making effective use of its compute resources is a critical management challenge. Similar issues have been pointed out at xAI, and the trend of companies with massive AI infrastructure turning into cloud providers is accelerating.

Meta's Non-Invasive Brain-Machine Interface "Brain2Qwerty v2" Closes In on Implants Without Surgery

Brain2Qwerty v2, developed by Meta's AI research division FAIR, is showing major progress in non-invasive (surgery-free) brain-machine interface technology.

The system reads magnetic signals from outside the skull and reconstructs what a person is typing. Restoring text from brain signals previously required surgery to implant electrodes in the brain, but Brain2Qwerty v2 is reportedly approaching the accuracy of implanted implants without any surgical procedure. Clinical applications for patients with paralysis remain a challenge for the future, but the possibilities of the non-invasive approach in the field of BCI (brain-computer interface) have expanded significantly.

Anthropic Adds New Security Measures, Government Restrictions on Its AI Models Lifted

After Anthropic introduced new security measures aimed at regaining the trust of the Trump administration, government restrictions on the company's "Fable 5" and "Mythos 5" AI models were lifted.

Restrictions on AI models by government agencies are sometimes imposed out of national security concerns. Anthropic addressed those concerns by strengthening its security standards and, as a result, won the lifting of the restrictions. The case is drawing attention as an example of how the relationship between AI companies and governments is changing through demonstrated safety.

Big Tech Starts Taking "Do Chatbots Have Feelings?" Seriously

According to a Washington Post report, the world's largest tech companies have begun seriously engaging with a question long relegated to the realm of science fiction: whether chatbots can have emotions or consciousness.

It has long been known that AI models can surface responses that appear emotional, but there is still no explanation for whether that is mere pattern reproduction or corresponds to some form of inner experience. When it comes to AI safety and ethics, this question is becoming a theme that can no longer be avoided.

Uber Overhauls the Leadership of Its AI Data Labeling Business

According to Bloomberg, Uber has significantly reshuffled the leadership of its AI data labeling business. Data annotation (labeling), which is essential for training AI models, occupies an important position in the AI value chain, but the industry is undergoing structural change amid cost competition and a wave of automation.

Uber's organizational shake-up is seen as a move reflecting intensifying competition in the data labeling market and the advancement of AI-driven labeling automation.

The Gap Between Open and Closed Models May Be "Smaller Than You Think"

According to a discussion gaining traction in Reddit's LocalLLaMA community, the performance gap between closed (commercial) models and open models may be smaller than generally believed.

The reasoning: benchmark results for closed models may include additional processing beyond model inference itself, such as post-inference corrections and system-level optimizations. The argument that when Claude outperforms GLM-5.2 on a benchmark, we cannot distinguish whether it reflects a difference in pure model capability or in the surrounding systems offers an important perspective for the open AI community.

---