Google and Amazon Are Betting Everything on Anthropic
Google is committing up to $40 billion to Anthropic — the largest AI investment in history — as Amazon simultaneously piles on a fresh $5 billion with up to $20 billion more to follow. Anthropic is pledging over $100 billion to AWS over the next decade. Meanwhile, Google Cloud Next '26 reveals 8th-generation TPUs and confirms 75% of all new Google code is now AI-generated. OpenAI's Sam Altman apologizes to the Canadian town of Tumbler Ridge, where a school shooter had described violent plans to ChatGPT months before the attack. And the DOJ joins Elon Musk's xAI in suing Colorado over its AI consumer protection law.
Google commits up to $40B to Anthropic. Amazon adds $5B more. A single AI lab just became the most capitalized bet in tech history.
In back-to-back announcements that landed this week, Google is investing $10 billion in Anthropic immediately, with the ability to invest up to $30 billion more based on performance targets — and Amazon, which had already invested $8 billion in Anthropic, added a fresh $5 billion on Monday with commitments of up to $20 billion additional in the future. Combined, two of the world's largest technology companies have now pledged a combined maximum of roughly $65 billion to a single AI startup founded just three years ago. Anthropic's side of the deal is equally staggering: the company has committed more than $100 billion over the next ten years to AWS technologies, spanning Graviton and Trainium2 through Trainium4 chips. The announcement from Anthropic describes the partnership as securing "up to 5 gigawatts of capacity" for training and deploying Claude — enough to power a small city, and a signal of the raw compute scale that frontier AI labs now require to stay competitive.
The Amazon deal also deepens Claude's distribution reach. Anthropic is launching the full Claude Platform on AWS, giving organizations direct access to Claude through their existing AWS accounts with no additional credentials or contracts required. Anthropic notes that Claude already runs on all three of the world's largest cloud platforms — AWS (Bedrock), Google Cloud (Vertex AI), and Microsoft Azure (Foundry) — a distribution footprint no other frontier lab matches. Over 100,000 customers now run Claude on Amazon Bedrock, a number that Dario Amodei called "increasingly essential to how they work" in his statement. The investment rounds cap a week in which it became undeniably clear that the hyperscalers are not hedging between AI labs — they are all-in on Anthropic specifically, despite simultaneously backing OpenAI through Microsoft.
anthropic.com ↗Google Cloud Next '26: 8th-gen TPUs land, 75% of Google's code is now AI-generated, and the agentic enterprise has a new mission control
Google's annual cloud conference delivered a burst of announcements on Thursday that set the infrastructure bar for the next generation of AI deployments. The headliner: Google's 8th-generation Tensor Processing Units, arriving in two configurations. The TPU 8t, optimized for training, scales to 9,600 TPUs in a single superpod with 2 petabytes of shared high-bandwidth memory — delivering three times the processing power of the previous Ironwood generation and up to 2x better performance per watt. The TPU 8i, optimized for inference, connects 1,152 TPUs in a single pod with 3x more on-chip SRAM, designed specifically to handle the massive concurrency demands of running millions of agents simultaneously at low latency. Both will be available to Google Cloud customers alongside its continuing NVIDIA GPU portfolio. Google also unveiled the Gemini Enterprise Agent Platform, described as "mission control for the agentic enterprise" — a full-stack system for building, scaling, governing, and optimizing AI agents across an organization.
The most striking data point from the conference came from Sundar Pichai himself: 75% of all new code written at Google is now AI-generated and approved by engineers, up from 50% just last fall. The pace of that shift — from half to three-quarters in roughly six months — is remarkable, and the context matters: Google is also moving toward "truly agentic workflows" in which engineers orchestrate autonomous coding agents rather than writing code directly. Pichai noted that Google Cloud's first-party models now process more than 16 billion tokens per minute via direct API use, up from 10 billion last quarter. The conference also confirmed that Apple's Gemini-powered Siri upgrade is still on track for "later this year," with Google teasing the partnership on stage with a giant Apple logo.
blog.google ↗OpenAI's Sam Altman apologizes to Tumbler Ridge — after it emerged the company knew about the shooter months before the attack
OpenAI CEO Sam Altman issued an apology to the Canadian town of Tumbler Ridge, British Columbia, this week, following a report revealing the extent of what OpenAI knew about the Tumbler Ridge Secondary School shooter before the February 10th attack that killed nine people and injured 27. Jesse Van Rootselaar, the suspect, had conversations with ChatGPT last June involving descriptions of gun violence that triggered the platform's automated review system. Multiple OpenAI employees raised internal concerns that the posts could be a precursor to real-world violence and urged company leaders to contact law enforcement. Leadership ultimately declined, concluding the activity did not constitute an "imminent and credible risk" of harm. OpenAI banned the account, but took no further action. OpenAI spokesperson Kayla Wood told The Verge that the company has now "proactively reached out to the Royal Canadian Mounted Police with information on the individual and their use of ChatGPT" and will support the investigation. Wood added that OpenAI's goal is to "balance privacy with safety and avoid introducing unintended harm through overly broad use of law enforcement referrals."
The case raises questions that extend well beyond OpenAI's specific decision. No clear legal or regulatory framework currently requires AI platforms to report concerning user behavior to law enforcement — a gap that exists in sharp contrast to industries like banking (mandatory suspicious activity reports) and healthcare (mandatory reporting requirements for threats of violence). The privacy-versus-safety tension OpenAI invoked is real: any reporting standard that's too broad would transform AI chatbots into surveillance tools and deter people from seeking help for mental health struggles through platforms they might otherwise use. But the Tumbler Ridge case is a hard example in the opposite direction, and Altman's apology — offered after the fact — doesn't address what standard OpenAI uses going forward or whether that standard will be applied consistently.
theverge.com ↗The DOJ joins xAI in suing Colorado over its AI discrimination law — and the constitutional argument is a preview of every AI regulation fight to come
The U.S. Department of Justice has joined an existing lawsuit filed by Elon Musk's xAI against Colorado's Consumer Protections for Artificial Intelligence law, set to take effect June 30th. The Colorado law requires AI developers to take "reasonable care to protect consumers" from algorithmic discrimination — a standard designed to prevent AI systems from producing outputs that discriminate against people based on protected characteristics. In its filing, the DOJ argues that the requirement violates the Equal Protection Clause of the Constitution, a framing that surprised many legal observers given that the Equal Protection Clause is typically invoked to protect individuals, not corporations. The move marks a significant escalation: a federal administration intervening in a state AI regulation battle on the side of an AI company, and doing so with a constitutional rather than a preemption argument.
Colorado's law is one of the first state-level AI consumer protection statutes to approach enactment. Its passage last year sparked a significant industry lobbying campaign, and it has become a test case for whether states can regulate AI applications independently or whether federal preemption — explicit or implicit — will render state AI laws unenforceable. The DOJ's intervention, combined with the administration's existing executive orders rolling back Biden-era AI safety requirements, signals a consistent federal posture: the U.S. government under the current administration intends to block sub-federal AI regulation that it views as economically constraining, regardless of whether a federal alternative exists to fill the gap.
theverge.com ↗Today's brief has a through-line that's easy to miss if you read each story in isolation: the AI industry is simultaneously consolidating its capital, scaling its infrastructure, failing its safety obligations, and pre-empting the regulation that might fix the latter. Those aren't unrelated trends. They're the same story told from four angles.
The Google/Amazon/Anthropic capital story is the obvious headline, and it deserves the attention it's getting. But the number that stayed with me from today's news isn't $65 billion — it's six months. That's how long it took Google's AI-generated code share to go from 50% to 75%. The $65 billion figure describes where capital is going. The six-month trajectory describes how fast the transformation of knowledge work is actually moving at the most sophisticated organizations on earth. Both numbers should inform how urgently people outside those organizations are thinking about what their own work looks like in three years.
The Tumbler Ridge story and the DOJ/Colorado story belong together. Tumbler Ridge is the consequence of an AI safety framework built on judgment calls and internal escalation processes that failed when it mattered most. The DOJ/Colorado story is the federal government making it harder for the one level of government that was actually trying to create external accountability — the states — to do so. You don't have to believe Colorado's specific law was perfectly drafted to believe that the gap between AI capability and AI accountability is widening, not closing, and that the current direction of federal AI policy is making that gap worse. At some point, the cost of that gap gets measured in ways that are harder to apologize for than a statement to The Verge.