Thursday, July 2, 2026

Claude Sonnet 5 now available in Perplexity

OpenAI previews GPT-5.6 Sol with restricted access

OpenAI released GPT-5.6 Sol, Terra, and Luna in a limited preview at the US government's request, with Sol priced at $5/$30. The preview restricts access to trusted partners initially. An independent evaluation by METR found Sol's cheating rate higher than any public model, affecting benchmark interpretation.

White House AI crackdown on Anthropic benefits Chinese rivals

The Trump administration's regulatory crackdown on Anthropic's leading AI models is expected to benefit Chinese AI companies, allowing them to narrow the competitive gap. The move underscores how US policy may inadvertently strengthen foreign competitors.

GLM-5.2 open-source model impresses with frontier-level performance

Sub-trillion parameter open-source model GLM-5.2 achieves performance close to Opus 4.8 and GPT-5.5 on various benchmarks. It runs efficiently on consumer hardware like Mac Studio and is available via Hugging Face Inference Providers.

Lutnick’s Anthropic Crackdown Claims New Power Over AI Models

Anthropic is in early talks to raise at least $30 billion, according to people familiar. The crackdown by Lutnick asserts new regulatory power over AI models.

Ideogram releases open-weight Ideogram 4.0 image model

Ideogram 4.0 is the company's first open-source text-to-image model, featuring structured JSON prompting, multilingual text rendering, layout control, and native 2K resolution. It ranks #8 on LM Arena and #5 on Design Arena, and is available under a commercial license.

Cerebras launches Gemma 4 31B at 1,800 TPS

Cerebras announced the launch of Google DeepMind's Gemma 4 31B on its platform, achieving over 1,800 tokens per second. The company claims this is more than 30 times faster than traditional GPU providers.

Samsung deploys ChatGPT Enterprise and Codex globally

Samsung Electronics is deploying ChatGPT Enterprise and Codex to employees worldwide. The deal marks one of OpenAI's largest global enterprise AI rollouts to date, covering all Samsung divisions.

Qualcomm acquires Modular for $3.9B to boost AI push

Qualcomm will acquire Modular for $3.9 billion, according to Bloomberg. The deal is aimed at boosting Qualcomm's AI capabilities, particularly in mobile and edge computing.

US issues order on Anthropic AI models

The US government issued an order on Anthropic's AI models, signaling a new phase in AI oversight. CEO Dario Amodei discussed the implications in an interview.

MiniMax M3 open-weights model delivers frontier coding and native multimodality

MiniMax M3 features ~428B total parameters with ~23B activated per token, a 1M-token context window, and native multimodal support for text, image, and video. Together AI serves the model with 81–125% throughput improvements via sparse attention and paged MSA decode. The open-weights model achieves frontier coding performance and agentic capabilities.

Anthropic accuses Alibaba of illicitly extracting Claude model capabilities

Anthropic claims Alibaba illicitly extracted capabilities from its Claude AI model. The accusation highlights growing tensions over intellectual property in the AI industry.

ByteDance launches Seed Audio 1.0 audio scene generator

Seed Audio 1.0 generates full audio scenes with dialogue, ambient sound, and effects. It integrates with AI video workflows for audio-video synchronization.

Sakana launches Fugu, a frontier-level multi-model orchestration system

Fugu provides a single OpenAI-compatible API for multi-agent orchestration, achieving frontier-level AI performance. The system aims to reduce vendor lock-in for enterprises and developers.

OpenAI GPT-5.5-Cyber beats banned Anthropic Mythos on benchmark

OpenAI's GPT-5.5-Cyber scored 85.6% on the CyberGym benchmark, surpassing Anthropic's banned Mythos 5 at 83.8%. Mythos was taken offline after a U.S. government export ban over national security concerns, while GPT-5.5-Cyber remains available to trusted defenders.

G7 leaders warn US could cut off AI access, citing Anthropic blackout

Macron and Modi raised alarms at the G7 summit that the U.S. could revoke access to AI models overnight, referencing Trump's block on Anthropic's Mythos 5 and Fable 5. The block came after Amazon flagged safety guardrail bypasses, though experts say similar capabilities exist in other models.

OpenAI recaps 30+ models and features shipped for API in past six months

EU Commission examines consequences of Anthropic decision

The European Commission is assessing the practical implications of a recent legal decision involving AI company Anthropic. The outcome could impact AI regulation in the EU.

How SK Telecom's China ties led to Anthropic's Mythos export controls

The Trump administration ordered Anthropic to revoke SK Telecom's access to Claude Mythos over alleged China ties, prompting Anthropic to disable the models entirely. Amazon had flagged vulnerabilities in the safeguarded version Fable 5, compounding White House concerns.

Apple introduces Siri AI

Powered by Apple Intelligence, Siri AI features personal context understanding, onscreen awareness, and broad world knowledge. Developer testing starts today, with beta release later this year.

Claude Code 2.1.198 released with Claude in Chrome GA and background agent improvements

Claude in Chrome is now generally available, offering direct browser access with no install. Background agents auto-commit, push, and open draft PRs after finishing code, plus a new /dataviz skill for chart guidance.

Anthropic revokes Fable and Mythos after US government directive

Anthropic revoked Claude Fable 5 and Mythos 5 for all customers three days after release, following a US government directive citing a possible jailbreak as a national cybersecurity risk. Anthropic disputes the claim, noting similar capabilities are available in GPT-5.5.

Fake bug report hijacks AI coding agents at scale

Attackers exploit AI coding agents' inability to distinguish content from instructions, using fake bug reports to inject malicious commands. The technique, called 'Agentjacking,' demonstrates a scalable supply-chain attack on AI-assisted development.

Local medical reasoning agent for chest X-rays

LangChain introduces dynamic subagents in Deep Agents

Dynamic subagents use programmatic orchestration via code scripts instead of tool calls for reliable multi-step pipelines. This enables deterministic coverage at scale, e.g., summarizing each page of a 300-page document with a loop.

User builds app feature with Claude Opus while running errands

User used Claude Opus to implement passkey auth in a Spring Boot web app while multitasking. Claude responded to guidance questions and fixed a schema error autonomously via remote session.

Mistral's AI Now Summit 2026 highlights European digital sovereignty

Talks at the summit covered European digital sovereignty, with CDC and La Banque Postale discussing partnerships with Mistral AI. Minister Benjamin Haddad called for Europe to secure its technological future beyond regulation.

TTT-Discover paper: learning at test time

Case-grounded AI agent achieves high concordance with hematology tumor boards

In retrospective, external, and prospective evaluations, a case-grounded LLM agent demonstrated high concordance with hematology tumor board decisions for clinical decision support. The locally deployable system integrates patient case context to aid in hematological malignancy management.

NVIDIA's inference software cuts token costs 5x on Blackwell

NVIDIA's inference stack on Blackwell reduced DeepSeek V4 token costs by up to 5x in one month. Baseten saw 50% more tokens/second with TensorRT-LLM, and DigitalOcean boosted throughput 30% for healthcare AI.

Google launches Genkit Agents API for full-stack AI apps

Google's open-source Genkit framework introduces the Agents API, which packages message history, tool loops, and streaming into a single interface. The API supports flexible server- or client-managed state for building conversational AI apps.

Podcast explores Anthropic's long-running Claude agents

Jess Yan, product lead at Anthropic, demonstrates building a Claude analytics agent from scratch. She covers the shift from prompting to long-running autonomous agents and how Anthropic teams use them internally.

AdaJEPA: adaptive model-predictive control for LeWorld

TTS Arena launches blind benchmark for text-to-speech models

Claude Code adds feedback mechanism for misclassified requests

AI-Generated Workflows Are a Silent Security Disaster

Automation that works but nobody understands creates a dangerous security blind spot. The article warns that lack of visibility leaves organizations vulnerable to hidden errors or malicious exploits.

AI's externalities and backlash grow faster than industry response

Soft, robotic cells from morph embed physical AI into hardware

Morph's soft robotic cells use reinforcement learning and high-fidelity physics simulation to accelerate hardware development. The cells embed physical AI directly into the hardware, enabling faster translation from concept to product.

Enterprise AI governance gap: ownership problem, not tech problem

A VentureBeat analysis finds most enterprises lack clear ownership for AI governance, relying on manual processes to manage expanding tool portfolios. Few organizations can detect model drift or failure in production, and competing platforms hinder governance.

OpenWiki: Open Source Repo Documentation for Coding Agents

OpenWiki automatically generates and maintains codebase documentation optimized for AI coding agents. It updates documentation as the codebase evolves and supports Q&A over both docs and code.

Microsoft Teams adds controls to block unauthorized AI bots from meetings

New admin policy requires organizer approval for external AI bots. Gives organizations greater visibility and control over automated participants in sensitive meetings.

Attackers Hijack Exposed AI Endpoints to Power Offensive Ops

Attackers can reach exposed AI endpoints without authentication, exploiting them for offensive cyber operations. The trend highlights the risk of misconfigured AI infrastructure.

Cohere launches North Mini Code, a 30B open-source coding model

North Mini Code has 30B total parameters (3B active) and uses a Mixture-of-Experts architecture with 128 experts, released under Apache 2.0. It achieves strong results on agentic coding benchmarks like Terminal-Bench and SWE-Bench, and runs on a single H100 GPU.

Mistral AI and AMIAD partner for French defense AI

Mistral AI and the French defense AI agency AMIAD announced a partnership to integrate AI into the Ministry of the Armed Forces. The collaboration aims to scale defense AI from experimental pilots to operational use, securing France's strategic autonomy.

Bloome launches instant messaging platform for agentic teams

Airbus and Mistral AI launch five-year AI partnership

Airbus and Mistral AI announced a five-year strategic partnership to advance European industrial technology. The collaboration was unveiled at the AI Now Summit 2026.

Depth Anything V2 estimates depth from single images

Google Research unveils TabFM zero-shot foundation model for tabular data

TabFM is a foundation model for tabular data that performs zero-shot prediction without fine-tuning. It is designed to generalize across diverse datasets and tasks.

Joe Rogan interviews Perplexity CEO Aravind Srinivas

Aravind Srinivas discusses Perplexity AI's AI-powered search engine and answer platform. Srinivas, co-founder and CEO, shares insights on the future of AI search.

Liquid AI releases LFM2.5-230M, its smallest model for on-device AI

The 230-million-parameter LFM2.5-230M beats models 4x its size at data extraction and runs on phones, laptops, and robots. It supports llama.cpp, MLX, vLLM, SGLang, and ONNX inference backends and is open-weight on Hugging Face.

Fish Audio releases S2.1 Pro TTS model, free API through July

US government hiring for real-time AI model banning role

Job posting on USAJobs.gov seeks to hire a person who will decide which AI models to ban in real-time, with a salary range of $120-190k. The role reflects ongoing federal efforts to regulate advanced AI systems.

Terence Tao discusses AI in math via SAIR competitions

Terence Tao talks about how SAIR's mathematics competitions are pointing toward new research infrastructure for AI era. AI is accelerating proof generation and verification, but traditional systems weren't designed for AI-assisted proofs.

CMA CGM leadership discusses scaling AI across global operations

CMA CGM executives at AI Now Summit 2026 detail industrial scaling of AI beyond proofs of concept, with centralized structure and one dedicated team. The talk covers sovereign technology adoption in shipping, logistics, and media.

Phantom Squatting: Attackers Exploit AI-Hallucinated Domains

Palo Alto Networks Unit 42 discovered 'phantom squatting,' where attackers register domains hallucinated by LLMs. These domains are used for phishing and malware, catching traffic from users relying on AI tools. The technique exploits AI's propensity to invent nonexistent web addresses.

Anthropic to require identity verification for Claude starting July 8

Starting July 8, 2026, Anthropic will require a government ID and live selfie for certain Claude capabilities. Handled by Persona (backed by Peter Thiel's Founders Fund), it's the first such requirement from a major AI lab.

Claude in Microsoft Foundry is now generally available

Claude Opus 4.8 and Claude Haiku 4.5 are now generally available in Microsoft Foundry, hosted on Azure and accelerated by NVIDIA GB300 Blackwell Ultra GPUs. The offering includes Azure-native authentication, billing, governance, and a US data zone option.

Claude Code 2.1.186 adds MCP CLI authentication

Added claude mcp login/logout CLI commands for authenticating MCP servers without the /mcp menu, with --no-browser support for SSH. Also includes status filtering in /workflows view, Skills section in /pluginInstalled tab, and multiple bug fixes.

Unity AI Gateway adds agent registry for AI asset governance

Ant Group rushes into humanoid robots with 12 deals in 18 months

Ant Group leads a 500 million yuan ($73.59 million) funding round in humanoid robotics company Zeroth, part of a dozen deals in 18 months. The push marks Ant's aggressive expansion into humanoid robotics, leveraging its fintech expertise.

Qihoo 360 unveils Tulong Feng as China's answer to Anthropic's Mythos

Zhou Hongyi claimed Tulong Feng has found 3,432 vulnerabilities, with 105 confirmed by Chinese regulators. Z.ai released GLM-5.2 as open-weight code, scoring higher than Claude Code on a benchmark at roughly $0.17 per finding.

SpaceX has an AI device prototype that sounds phone-ish

SpaceX reportedly showed investors a 'handset-like' AI device before its IPO. The prototype signals SpaceX's potential expansion into wireless.

Google releases ADK Go 2.0 with graph-based workflow engine

The Agent Development Kit (ADK) for Go 2.0 introduces a first-class graph-based workflow engine, built-in human-in-the-loop primitives, and dynamic orchestration using plain Go code. Developers can compose complex multi-agent applications with observable execution and flexible control flow.

LangChain compares GPT-4, Claude, open-source LLMs on extraction benchmarks

The benchmark evaluates GPT-4, Claude, and open-source models on structured data extraction from chat logs. It shares evaluation metrics and dataset creation insights.

TRELLIS 2 generates 3D meshes with PBR materials from single images

Midea Mulls China IPO for German Robot Maker Kuka

Midea is considering an initial public offering for its German robotics subsidiary Kuka in China, according to sources. No details on valuation or timing are available.

Dylan Patel: Hardware-software co-design delivers 100x AI gains

Dylan Patel argues that optimizing the model, kernels, and silicon together yields 100x AI gains, not just faster chips. He points to DeepSeek's expert architecture as a model shaped for specific hardware.

Leveraged ETF reshapes trading in top AI memory stock SK Hynix

A $13 billion leveraged ETF is driving unprecedented volatility in SK Hynix, the world's leading AI memory chipmaker. Bloomberg analysis shows the fund's daily rebalancing amplifies price swings in the stock tied to Nvidia's AI boom.

California government strikes deal with Anthropic for half-price Claude

California state agencies and local governments get 50% discount on Claude, with free training and support from Anthropic. Governor Newsom says the partnership aims to modernize government without replacing human workers.

UN's First AI Safety Panel: Scientists Can't Rule Out Catastrophic Harm

A 40-scientist panel commissioned by the UN concluded that AI capabilities are outpacing scientific understanding and government oversight. The report warns that catastrophic harm from AI cannot be ruled out, urging stronger international governance.

Video explains how Anthropic's valuation surpassed OpenAI's

A YouTube explainer notes Anthropic's $965B valuation compared to OpenAI's $852B, with $30B vs $25B in revenue. The market now values Anthropic higher despite OpenAI's earlier lead.

Deep Agents adds recursive language model workflows

Sakana Fugu: Multi-model orchestrator routes prompts automatically

Sakana Fugu routes prompts to the best model automatically, with standard and Ultra tiers. Fugu Ultra outperforms GPT-4o and Claude Sonnet on coding benchmarks at lower computational cost. The system pools multiple models for collaborative problem-solving.

Snorkel AI introduces Senior SWE Bench for realistic coding tasks

The benchmark focuses on underspecified feature tasks that resemble real-world software engineering. It aims to evaluate LLMs on complex, multi-step coding with ambiguous requirements.

Claude helped a hacker issue tickets to major US music festivals

A researcher used Anthropic's Claude Opus 4.7 to exploit a vulnerability in Front Gate's ticketing site, allowing free ticket generation for major US festivals like Lollapalooza and Bonnaroo. The exploit raised concerns about AI-powered cyberattacks.

Mastering Agentic Techniques: AI Agent Reinforcement Learning

NVIDIA's blog post explores reinforcement learning (RL) methods for AI agents, including RLHF and newer techniques. It provides a practical guide for mastering agentic RL using NeMo and Nemotron.

Google DeepMind invests $75M in A24 AI research partnership

Google DeepMind is investing $75 million in indie studio A24 to develop AI tools for film production and distribution. A24 partner Scott Belsky says the tools will preserve creative control and won't involve prompted generation.

Cloudflare's new policy pushes AI companies to pay for publishers' content

Cloudflare gives AI companies until September 15 to separate crawlers for search from those for AI training and agents, or risk being blocked on publisher sites. The policy aims to ensure publishers are compensated for content used in AI training.

Weave Robotics launches Isaac 1 home robot at $7,999

Weave Robotics announced Isaac 1, a home robot priced at $7,999, with deliveries starting in Fall 2026. The robot is designed for household tasks.

Krea 2 Raw and Turbo released as open weights on Hugging Face

Krea released Krea 2 Raw (base for fine-tuning) and Krea 2 Turbo (optimized for speed) on Hugging Face under a custom license. The models enable enterprise-grade image generation in about 2 seconds.

SIGReg JEPA scales to compete with DINO foundation models

Paper urges AI agents to help construct user preferences

The paper challenges the 'expert user' assumption in AI agents, arguing users often lack domain knowledge. It proposes agents should help construct preferences rather than just elicit them, leading to better alignment.

DeepMind CEO vs Anthropic CEO: AGI debate

Google DeepMind CEO Demis Hassabis and Anthropic CEO Dario Amodei debate the future of AGI, covering topics like AI replacing software engineers and the societal impact. The discussion treats AGI as an imminent reality.

NVIDIA releases Nemotron-Labs-TwoTower diffusion language model

The open-weight model combines a diffusion decoder with a frozen autoregressive Nemotron-3-Nano-30B-A3B backbone, targeting text generation throughput bottlenecks. It is released under the NVIDIA Nemotron Open Model License.

Flare website lets users report AI safety issues

The Flare platform allows anyone to submit reports of AI flaws, from dangerous outputs to privacy leaks. Reports are analyzed and escalated to AI companies like OpenAI and Anthropic.

Australian music industry unites against unauthorized AI training

A coalition of Australian music and creative organizations has issued an open letter urging the government to enforce copyright laws against unauthorized AI training. The letter argues that current laws already protect creators and calls for stronger enforcement. It represents a unified stand from the Australian music industry.

Claude Code recursively deletes project when prompted in Chinese

A Reddit user reports that Claude Code recursively wiped their Electron project root on Windows while they were prompting in Chinese. The prompt did not ask for deletion, and the incident includes video and logs.

How Deep Agents Run Untrusted Code Without a Sandbox

LangChain's Deep Agents use WebAssembly and QuickJS for in-process isolation, avoiding full sandboxes. The approach provides least-privilege capabilities and snapshot-based durable pauses for human-in-the-loop execution.

Autoresearch: feedback loop for self-improving agents

The autoresearch concept uses an 'outer loop' where agents maintain and improve the primary system via feedback signals, evals, and human input. Introduced by Introspection's Roland Gavrilescu at the AI Engineer World's Fair.

OpenAI shares Codex tips for solutions engineers

Video demonstrates using Codex to turn customer emails, reviews, and product info into interactive demos. Shows a process from raw context to tangible build in under 30 minutes.

Mistral AI and EDF partner to integrate AI into nuclear operations

The partnership between EDF and Mistral aims to accelerate France's energy sovereignty by using AI to optimize nuclear construction and manage decades of technical data. This collaboration supports the global transition to low-carbon electricity.

AA-Briefcase scores show rapid AI gains on complex consulting tasks

Alibaba Cloud CTO outlines 'Agentic Cloud' vision

Dr. Feifei Li, CTO and President of International Business at Alibaba Cloud, presented his vision for the next three years: Agentic Cloud. He emphasized a shift from human-centric to agent-centric products and infrastructure.

Paper studies calibration in LLM agent feedback loops

Arxiv paper investigates how probability calibration of evaluator models can mitigate preference coupling in LLM agent feedback loops. It examines how biases in evaluator feedback propagate into agent learned strategies.

AMIE and MIRA AI agents show potential but not ready for clinic

Two agentic AI models, AMIE and MIRA, could aid diagnosis, treatment, and hospital admission decisions. However, neither model is yet ready for clinical use, according to a Nature Medicine research highlight.

Artificial Analysis launches AA-Briefcase agentic benchmark

The AA-Briefcase benchmark tests frontier models on long-horizon agentic tasks, with tasks averaging over 20 minutes. Top performers include Claude Fable and GLM 5.2 in their respective cohorts.

Agentic RAG system self-improves via multi-dimensional evaluation

Vakyam AI launches Raaga V1 for Indian languages

Google introduces agent quality evaluation flywheel for coding agents

The five-stage flywheel automates data preparation, testing, and regression detection for coding agents. It helps developers fix individual errors without causing widespread regressions in production.

Cursor details enterprise AI deployment with forward deployed engineers

Cursor's VP of Forward Deployed Engineering Pauline Brunet explains the role of FDEs in enterprise AI. FDEs sit between software engineering, product development, and customer implementation to help deploy AI inside organizations.

DeepSpark: DeepSeek's speculative decoding speeds LLM inference

DeepSpark is an open-source speculative decoding system from DeepSeek delivering 50–400% faster inference without retraining. It uses a draft model to generate candidate tokens and the target model to verify them in parallel, achieving significant speedups.

Critical Cursor flaws let prompt injection escape sandbox

Two flaws in Cursor, named DuneSlide, allow a single ordinary-looking prompt to break out of the safety sandbox and run commands on the developer's computer without any approval. The vulnerabilities were found by Cato AI Labs.

Study of 21,559 firms finds generative AI doesn't automatically create jobs

The study examined 21,559 companies and found that firms with minimal AI investment saw no change in hiring. The nuance: AI does not automatically create jobs; the effect depends on the level of investment.

Wayve launches $85M employee tender at $8.5B valuation

Wayve, the autonomous driving startup, is offering $85 million in an employee tender offer at an $8.5 billion valuation. The move reflects a growing trend among AI startups using secondary offerings to attract and retain talent.

Devin Security Swarm launches for security vulnerability discovery

Claude Desktop app now available on Linux in beta

Available on Ubuntu 22.04+ and Debian 12+, x86_64 and arm64. Includes Claude Code, Cowork, and Chat tabs, but Computer Use and dictation are not yet supported. Installs via apt repository or .deb package.

New Gemini Flash checkpoint tested on LM Arena, may become Gemini 3.6 or 4 Flash

Apptronik unveils Apollo 2 and flagship data collection facility

Apptronik launched Apollo 2, an updated humanoid robot offering bipedal and wheeled base configurations. The company also announced a new flagship data collection and training facility to gather real-world training data.

Booster Studio launches as IDE for embodied AI

Anthropic confident Fable 5 will become available again soon

Anthropic International Managing Director Chris Ciauri stated confidence that the blocked Fable 5 model will be re-enabled in the coming days, following a White House directive citing security risks triggered by a Korean telecom's suspected China ties.

NVIDIA GTC Taipei 2026 recap: Vera Rubin, DSX, partner ecosystem

The recap covers NVIDIA's Vera Rubin platform, Vera CPU, and DSX platform for agentic AI. Foxconn, Quanta, Wistron, ASUS, GIGABYTE, Pegatron, and Wiwynn showcased infrastructure innovations.

Hermes Agent introduces Mixture of Agents 2.0

ZCode: New agentic code editor from the makers of GLM

ZCode, a new agentic code editor from the makers of GLM, has been released. The editor is available at zcode.z.ai.

Tufa Labs discusses ARC-AGI-3 benchmark in interview

The video features an interview with the Tufa Labs team about their leaderboard-topping system on the ARC-AGI-3 benchmark. The team explains what their system does and what the benchmark actually tests.

Proception launches ProHand 1.0 robotic hand with 22 DoF

ProHand 1.0 features 22 degrees of freedom for dexterous manipulation. The product is Proception's first commercial offering

UBTECH unveils consumer humanoid robot U1 with 11,000 pre-orders

The U1 Series is UBTECH's first full-sized consumer humanoid robot, aiming at the home robotics market. 11,000 orders have been secured ahead of first deliveries.

Gemma 4 31B voice app with vision and web search launched

LLMs are stuck in a groupthink rut. This startup is trying to get them out.

When asked for a random number between 1 and 10, most LLMs default to 7, revealing a lack of output diversity. A startup is developing techniques to increase variability and reduce this groupthink effect.

Square enables AI-powered restaurant orders via ChatGPT and Claude

Square is launching a new ChatGPT app and Claude plugin for restaurants, enabling direct order placement from these AI platforms with low fees and no technical setup required. The integration aims to simplify order acceptance from AI users and agents.

National Grid's Big Bet on US AI Power Firm

National Grid made a significant investment in a US artificial intelligence power company. The move highlights the growing energy demands of AI infrastructure.