Original AI News | Winzheng

Three US States Tighten AI Data Center Approvals in Succession; 474 GW of Power Applications Triggers Infrastructure Governance Crisis

Three US states have moved within six weeks to restrict AI data center expansion amid a 474 GW backlog of grid interconnection requests in Texas and more than $1 trillion in unfulfilled lease commitments from tech giants. The regulatory shift — from New York's statewide moratorium to Pennsylvania's legally binding GRID framework — reflects a structural struggle over who bears the costs of AI's energy and infrastructure demands.

China's Ten Departments Jointly Implement AI "Review-Before-Creation": What the World's First Mandatory Pre-Research Ethics Review Mechanism Means

On April 2, 2026, ten Chinese government departments jointly issued the "AI Science and Technology Ethics Review and Service Measures (Trial)," requiring ethics review before AI research begins — a world-first approach that marks China's shift from policy release to compliance enforcement in AI governance. The mechanism's timing fundamentally diverges from the EU AI Act's pre-market compliance model.

EU AI Office Sends First RFIs to OpenAI and Others on August 29, Formally Launching Mandatory Enforcement Phase

On August 29, 2026, the EU AI Office issued its first information requests to general-purpose AI model providers including OpenAI, Anthropic, and Google, covering model safety, independent external evaluation, post-deployment monitoring, and training data summary disclosure. The move marks the formal start of the mandatory enforcement phase.

DeepSeek Raises 50 Billion Yuan at $74 Billion Valuation, Begins Preparations for 2027 STAR Market IPO

DeepSeek is close to completing a funding round of approximately 50 billion yuan at a valuation of about 500 billion yuan ($74 billion), expected to close by the end of August. The Hangzhou-based AI company has also initiated preparations for a 2027 STAR Market listing.

Research Lab

4-Model Translation Showdown: Week 36 Quality Assessment, claude-sonnet-4.6 Leads with 9 Points

This week, 405 translation tasks were completed by 4 models. In a sampled blind comparison of 3 articles across multiple models, claude-sonnet-4.6 ranked highest overall with an average score of 9/10.

Grok 4.7 Training Complete but Delayed to September: The SpaceX Gamble Behind 2.1 Trillion Parameters

Grok 4.7's initial training is complete, but xAI has delayed its release to early September to inject SpaceX engineering data—including Starlink telemetry, rocket test logs, and internal engineering documents—into the model. With approximately 2.1 trillion parameters, Grok 4.7 marks a 40% scale increase over Grok 4.6.

Grok 4 Leads with 89.15: 2026-08-31 Smoke Quick Test Data Brief

On 2026-08-31, the YZ Index Smoke quick test covered 11 models, with Grok 4 ranking first at 89.15 points. Smoke is a daily 10-question quick test suited for short-term signals, not equivalent to the Full weekly ranking conclusions.

a16z Establishes $1.1 Billion Machine Age Fund Focused on AI Physical Infrastructure

On August 28, 2026, a16z announced a $1.1 billion Machine Age Fund dedicated to investing in AI physical infrastructure, spanning chips, memory, data centers, cooling systems, robotics, and edge devices. The fund reflects the firm's strategic shift from software toward the physical layer amid surging AI compute and memory demands.

Amazon Shuts Down Mechanical Turk After 21 Years: The AI It Helped Train Ultimately Replaced It

Amazon has announced it will shut down the Mechanical Turk crowdsourcing platform on September 30, 2026, along with companion services SageMaker Ground Truth and Amazon Augmented AI. The move marks the end of a human data infrastructure era—the AI that MTurk helped train ultimately learned to perform the very tasks it once assigned to human workers.

All Three Music Giants Move: Sony and Warner Sue Anthropic as Copyright Battle Enters Decisive Phase

Sony Music Publishing and Warner Chappell Music have filed a joint lawsuit against Anthropic and its founders, joining Universal and other plaintiffs in a copyright fight that could theoretically reach billions of dollars in damages. With all three major music publishers now aligned against the AI company, the case signals a decisive phase in the industry's battle over training data.

Tencent Open-Sources Hy4 Preview: 770B-Parameter Model Outperforms DeepSeek V4 Pro on Coding Benchmarks, but Over-Verification Slows Real-World Delivery

Tencent Hunyuan has open-sourced Hy4 Preview, a 770B-parameter MoE model that scores 85.4 on Terminal Bench 2.1—surpassing DeepSeek V4 Pro and tying with Claude Opus 5—while its DeepSWE score jumps from 28.0 to 64.3. However, the model's over-verification tendency prolongs response times, and visual input support remains absent in this preview version.

White House AI Legislative Framework: Regulatory Sandboxes First, Federal Uniformity Over 40 States

On March 20, 2026, the White House released the "National AI Policy Framework: Legislative Recommendations," a non-binding document prioritizing regulatory sandboxes, developer liability protection, and federal preemption of fragmented state AI laws. Welcomed by major industry groups, the framework faces sharp divisions in Congress.

NVIDIA Acquires Hugging Face for $12.9 Billion: Why the Chip Giant Is Willing to Pay an 86x Revenue Premium

NVIDIA has agreed to acquire open-source model platform Hugging Face for approximately $12.9 billion, marking one of the largest acquisitions in the chip giant's history. If finalized, the deal would bring the global AI developer community's most important model hub under the control of a chip company.

Gemini 3.1 Pro Main Score Plunges 10.8 Points; Material Constraint Drops 14.4 in a Single Day

Gemini 3.1 Pro's main score in today's Smoke evaluation fell from 96.60 to 85.83, a drop of 10.8 points, driven primarily by a 14.4-point decline in the material constraint dimension.

Grok 4 Main Ranking Plunges 8.8 Points: Code Execution Drops from 100 to 90.7

Grok 4's main ranking score in today's Smoke evaluation fell from 96.99 to 88.14, a single-day decline of 8.8 points, with code execution dropping from 100.00 to 90.70. The decline appears driven by question-draw fluctuation rather than genuine model degradation.

Claude Opus 4.7 Trends Up 10.1 Points, Gemini 3.1 Pro Falls 11.2 Points: 2026-W35 Smoke Weekly Trends

In the Smoke evaluation from August 24 to August 30, 2026, Claude Opus 4.7 rose from 82.98 points on day one to 93.08 points on the final day, a trend increase of 10.1 points, while Gemini 3.1 Pro fell 11.2 points over the week.

Claude Opus 4.7 Tops with 93.08 Points: 2026-08-30 Smoke Quick Test Data Briefing

On 2026-08-30, the YZ Index Smoke quick test covered 11 models, with Claude Opus 4.7 ranking first with 93.08 points. Smoke is a daily 10-question quick test suited for observing short-term signals and is not equivalent to Full weekly ranking conclusions.

OpenAI Launches Rosalind Workbench Life Sciences Tool Platform, Competing with Anthropic

On August 28, 2026, OpenAI released the Rosalind Workbench research preview via the ChatGPT App, integrating the GPT-Rosalind model with bioinformatics tools. Amgen, Moderna, and the Allen Institute have joined as early collaborators.

Anthropic Opens Claude Enterprise Free to All US School Districts: One-Year Trial for Educational Data Moat

Anthropic has expanded Claude for Teachers into a free Enterprise offering for all US K-12 schools and districts, trading a one-year trial period for institutional embeddedness in the American education system. The move intensifies a three-way competitive race with Google and OpenAI while drawing scrutiny over FERPA compliance and data processing history.

Gates Proposes Robot Tax and Reserved Human Jobs: A Long-Overdue Policy Bill

In a 6,000-word essay on his personal blog, Bill Gates converts years of general statements into two concrete policy designs: taxing AI token calls and robot usage, and establishing a "Human Reserved" job category system. Written against the backdrop of AI becoming the leading cause of layoffs for five consecutive months, the proposal is a long-overdue effort to turn abstract debate into implementable mechanisms.