News

Nvidia says its Rubin generation reference design for a fully liquid-cooled AI data center runs hotter while eliminating “massive amounts of power usage” and “pretty much all water usage.” The claim matters because data center water and energy use has become a major public concern, but Nvidia still doesn’t address construction impacts, power-generation demands, or the cost versus air-cooled designs.

SpaceX has inked a compute deal with Reflection AI, which will pay $150 million a month starting July 1, 2026 through 2029 for immediate access to Nvidia’s latest GB300 AI chips and supporting hardware at SpaceX’s Colossus 2 data center near Memphis, Tennessee. The deal highlights how scarce top-end AI compute has become, with a long-term commitment worth about $1.8 billion a year to secure cutting-edge GB300 capacity.

AWS is in talks to sell its AI chips to other data centers, expanding beyond internal use as Amazon looks to challenge Nvidia more directly. CEO Andy Jassy has said the market could represent a $50 billion opportunity, highlighting how much revenue AWS thinks custom chips could generate.

FERC ordered grid operators to create a fast lane for data center interconnections, but it did not resolve the underlying shortage of electricity supply. The move could speed up AI infrastructure buildouts, yet without more generation it may simply shift bottlenecks from connection queues to power availability.

Amazon said its global data center operations consumed 2.5 billion gallons of water in 2025, or 0.12 liters per kilowatt-hour, down 2% from 2024 even as it expanded operations. The disclosure lands amid rising scrutiny of AI data centers’ water and power use, and Amazon says its efficiency is better than some Big Tech rivals.

OpenAI says a new report found PRC-linked influence operations using AI to target U.S. tech debates, data center narratives, tariffs, and false claims about ChatGPT. It matters because it shows state-linked actors are trying to shape AI policy and public opinion with generated content rather than just traditional propaganda.

Meta signed its first AI data center deal in India with Reliance for a 168-megawatt facility that will support Meta’s global AI computing needs and can be expanded over time. It marks an important infrastructure expansion for Meta in India and adds scalable capacity for training and serving larger AI workloads.

General Motors announced new vehicle-to-grid capabilities for its current EV and home energy customers, plus a commercial energy storage strategy built around newly developed sodium-ion batteries for industrial-scale grid applications and a new feature to simplify public charging. The move is meant to help offset rising electricity demand from AI data centers by turning idle EV batteries into distributed grid resources.

AirTrunk plans to invest $30 billion to build 5GW of AI data center capacity in India. The scale is notable because 5GW is a massive power footprint for AI infrastructure, signaling a major expansion of compute capacity in one of the world’s fastest-growing digital markets.

TSMC said demand from American customers is so high that even its U.S. factory buildout cannot keep up, with CEO C.C. Wei saying, “we can only support so much” and that the company is trying not to become a bottleneck. The AI boom is now straining semiconductor supply beyond memory chips like RAM and NAND Flash, highlighting how quickly demand for AI infrastructure is outpacing manufacturing capacity.

Google said it will aim to replenish more water than it uses at its data centers by 2030 and outlined five commitments in a new blog post, including investments in local water infrastructure, alternative water sources, and greater transparency around water use. The pledge matters because AI data center expansion has drawn backlash over resource consumption, and water has become a key pressure point for communities hosting the buildout.

Nvidia is aiming at the $200 billion CPU market by pushing AI agent PCs with Microsoft, Dell, and HP, betting that it has found a way to bring AI agents to consumers and enterprises easily, safely, and usefully. If successful, it would expand Nvidia beyond GPUs into mainstream personal computers and put it in direct competition with established CPU vendors.

OpenAI has broken ground on a 1 GW data center project in Michigan as part of Stargate. The build is meant to expand AI infrastructure, improve access, create jobs, and support local communities.

NVIDIA announced Cosmos 3, describing it as the first open omni-model for physical AI reasoning and action. It matters because an open model aimed at reasoning about and acting in the physical world could help accelerate robotics and embodied AI research.

Large exchanges are designing derivative products around AI tokens, treating them less as a computational output and more like a tradable input akin to electricity or bandwidth. This matters because it signals a shift toward financial markets pricing AI capacity itself, not just the models or services built on top of it.

Snowflake signed a five-year, $6 billion deal with Amazon Web Services to secure AI CPU chips for its cloud workloads. The agreement highlights continued demand for non-GPU AI infrastructure and adds to pressure on Nvidia as cloud customers diversify their chip suppliers.

Sundar Pichai said Google responded to ChatGPT by reworking its structure and leadership, and he discussed new Gemini models, AI agents, and major Search and YouTube changes after Google I/O. The notable implication is that Google is moving toward search that triggers tasks through the Gemini Spark agent platform, a shift that could accelerate “Google Zero” and further reduce traffic to the open web and creators.

Elon Musk’s xAI is reportedly leaning heavily on natural gas for its power needs, while SpaceX is focusing on orbital data centers instead of Earth-based solar power. The shift is notable because it contrasts with Musk’s earlier “solar-electric economy” rhetoric and suggests his companies are prioritizing practical compute infrastructure over renewable-energy promises.

Jensen Huang said Nvidia has found a “brand new” $200 billion market in CPUs for AI agents, marking a new growth target for the company. The prediction highlights Nvidia’s push beyond GPUs into AI infrastructure for agentic workloads, with Huang putting the opportunity size at $200 billion.

Box Elder County commissioners approved the Stratos Project, a 40,000-acre data center in Utah’s Hansel Valley backed by Kevin O’Leary and billed as a push for American AI dominance. It would be more than twice the size of Manhattan and could draw 9 GW of power, raising concerns about environmental damage and stress on already strained water supplies.

NVIDIA Cosmos Predict 2.5 is being fine-tuned with LoRA and DoRA for robot video generation. This matters because parameter-efficient adaptation can improve task-specific robot video outputs without full-scale retraining of the base model.

An interactive map built by Isabelle Reksopuro tracks data center construction and AI policy, prompted in part by Oregon residents’ confusion over claims that Google was taking public land for its facilities near The Dalles. The map helps separate local controversy from the legal and technical details, including a 150-acre Mount Hood National Forest land claim tied to municipal water access and a city of 16,010 people.

Gallup found that more than 70% of Americans oppose building AI data centers in their area, with only 7% saying they strongly favor them, based on March and April 2026 surveys of 1,000 randomly selected adults and 2,054 Gallup Panel members. The result is notable because Americans said they would rather live near a nuclear power plant than a data center, and opposition to data centers exceeded the 63% peak opposition ever recorded for nuclear plant construction.

xAI is reportedly operating nearly 50 gas turbines without clear oversight at its Colossus 2 data center in Mississippi, prompting a lawsuit over the company’s use of “mobile” turbines as power plants. The dispute matters because it highlights the environmental and regulatory scrutiny facing AI infrastructure as companies rapidly scale power-hungry data centers.

Google and SpaceX are reportedly in talks to build data centers in orbit, framing space as a potential future location for AI compute despite much higher costs than terrestrial infrastructure today. The idea is notable because it would extend AI infrastructure beyond Earth, but the economics and engineering hurdles remain substantial.

NVIDIA engineers and researchers are using Codex with GPT-5.5 to ship production systems and turn research ideas into runnable experiments. The notable detail is that the workflow spans both engineering and research, showing Codex being used not just for code generation but for moving ideas into runnable, production-oriented systems.

MachinaCheck is a multi-agent CNC manufacturability system built on AMD MI300X. It matters because it applies multi-agent AI to manufacturing validation, suggesting faster automated checks for whether parts can be machined.

Nvidia has already committed $40 billion to equity AI deals this year, continuing to expand its investment footprint across the AI ecosystem. The scale of that commitment underscores how central Nvidia has become to financing and shaping the industry around its GPUs, even as demand for AI infrastructure keeps rising.

SpaceX is planning to invest at least $55 billion in its “Terafab” AI chip plant in Austin, Texas, according to a public hearing notice tied to a tax-break request, with the total potentially reaching $119 billion if later phases are built. Musk originally said in March that the factory could eventually produce enough chips to support up to 200 gigawatts per year of compute, making it one of the most ambitious chip manufacturing plans in the U.S.

OpenAI introduced MRC (Multipath Reliable Connection), a new supercomputer networking protocol released through OCP to improve resilience and performance in large-scale AI training clusters. The protocol is aimed at large-scale AI training networks where reliability and throughput are critical, and its OCP release suggests broader hardware interoperability for cluster operators.

A timeline of GPT-5 “goblin outputs” tracks how a personality-driven quirk spread through model behavior, identifies its root cause, and outlines fixes. The notable detail is that the issue appears tied to model personality dynamics rather than a single isolated bug, which makes diagnosing and correcting it more complex.

OpenAI is scaling Stargate to build the compute infrastructure for AGI, adding new data center capacity to meet growing AI demand. The notable detail is that this is about expanding the underlying infrastructure rather than a specific model release, signaling a larger push to support future AI workloads.

Google released a new video explaining how its TPUs power increasingly demanding AI workloads. It highlights the role of TPUs in handling heavier AI training and inference demands, though the excerpt provides no specific performance numbers or model names.

The piece outlines 10 practical ChatGPT Codex use cases for everyday work, focused on automating tasks, creating deliverables, and turning real inputs into outputs across tools, files, and workflows. It matters because it frames Codex as a general workflow tool rather than just a coding assistant, highlighting how it can connect disparate work inputs into usable outputs.

Google is launching two specialized eighth-generation TPU chips for the “agentic era” of AI. The notable detail is that these are purpose-built accelerators, signaling a shift toward hardware optimized for AI agents rather than general-purpose model training alone.

OpenAI describes how to create and use ChatGPT skills for reusable workflows, automating recurring tasks, and producing consistent outputs. The notable detail is that skills are positioned as a way to standardize higher-quality results across repeated work without rebuilding prompts each time.

Custom GPTs let users build purpose-built AI assistants to automate workflows and produce more consistent outputs. They matter because they can tailor behavior to specific tasks, improving reliability and reducing repeated prompting.