Hut 8 Leases Texas AI Data Center for $9.8B — Crypto Miners' AI Pivot Accelerates
**Hut 8 Corp**, a major cryptocurrency mining company, has announced a **15-year lease worth a minimum of $9.8 billion (approximately ¥1.47 trillion)** for a new AI data center to be built at the "Beacon Point" campus in Nueces County, Texas. The tenant is described as a "high-investment-grade company," though its name has not been disclosed.
The deal could reach a total of **$25.1 billion** if the tenant exercises all three five-year extension options. Originally known for its Bitcoin mining operations, Hut 8 has been pivoting toward AI data center development, and this large-scale contract signals that the strategy is starting to deliver tangible results.
The shift from crypto mining to AI infrastructure has become an industry-wide trend, with companies accelerating efforts to leverage their existing large-scale power infrastructure to meet AI computing demand.
SpaceX Proposes $55B "Terafab" Semiconductor Plant in Texas
**SpaceX, led by Elon Musk**, has been revealed to be proposing the construction of a **next-generation semiconductor manufacturing facility called "Terafab"** in Grimes County, Texas. The first-phase investment amounts to **$55 billion (approximately ¥8.25 trillion)**, and the total could reach **$119 billion** once additional phases are completed.
According to publicly disclosed notification documents, Terafab is positioned as a "next-generation, vertically integrated semiconductor manufacturing and advanced computing manufacturing facility." SpaceX apparently plans to apply the manufacturing expertise it has cultivated through rocket development to the semiconductor field.
Domestic sourcing of semiconductors — a critical piece of the AI race — is a strategic priority for the United States, making SpaceX's entry a notable development for incumbent semiconductor manufacturers as well. That said, the specifics of the plan and its timeline remain to be seen.
ARIS: Different AI Models Collaborate Adversarially to Automate Research
**ARIS (Autonomous Research via Adversarial Multi-Agent Collaboration)**, featured on Hugging Face's Daily Papers, is attracting attention. ARIS is a framework that automates ML research workflows through **adversarial collaboration between different model families**.
Specifically, it uses Claude as the Executor and adopts a different model family, such as GPT, as the Reviewer. The design captures "correlated errors" that self-critique within a single model tends to miss by having different models check each other's work. It comes with 65+ Markdown-defined skills, 5 end-to-end workflows, and a three-stage evidence→claim audit cascade, and supports multiple executors including Claude Code, Codex CLI, and Cursor.
The project is already open source on GitHub, reportedly boasting over 8,000 stars, more than 5,000 users, and over 30 community-contributed skills. As a cross-model approach that compensates for the limitations of a single model, it could influence the future development of agentic research tools.
HuggingFace Open ASR Leaderboard Introduces Anti-Benchmark-Gaming Measures
HuggingFace has introduced **"Benchmaxxer Repellant"** (anti-benchmark manipulation measures) to its Open ASR (Automatic Speech Recognition) Leaderboard. To prevent fraudulent score manipulation on the leaderboard, the company has implemented measures such as adding evaluations on private data.
"Benchmaxxing" — over-optimizing models on test data to inflate scores — has long been a challenge in AI model benchmarking. Overfitting to public test sets has been a particular concern in the speech recognition field, and these new measures represent an important step toward maintaining the fairness of open evaluation.
Zed Team: "We're Not Building AI Features for the Money" — An Editor Company's Philosophy
The team behind the Zed code editor has published a blog post titled **"We're Not Building AI Features for the Money."** Highlighting the philosophy behind the company's AI feature development, the post emphasizes a commitment to fundamentally improving the developer experience rather than piling on monetization-driven AI features.
The post has sparked discussion on Hacker News, with a growing number of voices questioning the short-term "add AI features and it will sell" approach amid the AI boom. The stance of integrating AI while prioritizing the fundamental performance of the editor as a tool stands as a model for tool development.