April 30 Roundup: Anthropic eyes a $900B valuation, hyperscaler capex sets a record, Google signs the Pentagon, and the EU AI Act stalls

1. Anthropic opens talks at a $900B valuation — and would leapfrog OpenAI

Anthropic is in talks with investors to raise capital at a valuation of $900 billion, CNBC and Bloomberg reported on April 29. No term sheet has been signed and discussions are private, but the implications are enormous: the Claude maker would, for the first time, pass OpenAI on paper. OpenAI was last valued at $852 billion in late March after a $122 billion round that drew up to $50 billion from Amazon, $30 billion from Nvidia, and $30 billion from SoftBank.

Reuters, citing Bloomberg, reported that Anthropic is "weighing raising funds in a new round that would value the Claude maker at more than $900 billion." Just six weeks ago, Anthropic was a $380 billion company. Six weeks before that, it was running at roughly a $9B annualized revenue run-rate. As of this month, that run-rate is "$30 billion in annualized revenue," according to CNBC.

The trigger isn't just Claude usage. CNBC's source says the company "needs to be able to purchase the compute that it needs in order to run Mythos" — Anthropic's cybersecurity-focused model that has already pulled the Trump administration, the Federal Reserve, the Treasury, and major bank CEOs into closed-door briefings since early April. Anthropic's compute footprint now includes up to 5 gigawatts via Amazon (part of Amazon's up-to-$25B investment), another 5 gigawatts via Google–Broadcom that comes online in 2027, and Google's separately announced commitment to invest "up to $40 billion in Anthropic" — $10B now at a $350B mark, with $30B more if performance targets hit.

2. Hyperscaler AI capex sets a record — Alphabet alone spent $36B in the quarter

The four U.S. hyperscalers reported earnings into a market already nervous about AI returns. The result, per The New York Times: "A.I. spending sets a record, with no end in sight." Google parent Alphabet's spending was $36 billion in the quarter, more than double the $17 billion it laid out in the same period last year. Bloomberg reported Alphabet's Q1 revenue, excluding partner payouts, was $94.7B versus the $91.6B analyst consensus, with cloud revenue beating estimates as enterprise AI demand kept building.

The capex chart spooked some investors but didn't dent the spending plans. CNBC noted that the four hyperscalers were reporting "with a single company that doesn't even release its financials to the public" — OpenAI — looming over their results. The same day, AI-related stocks including Oracle and CoreWeave sold off after a Wall Street Journal report said OpenAI had recently fallen short of its own user-growth and revenue targets, even as it pushes toward an IPO. The selloff faded by close.

The other side of capex is monetization, and yesterday delivered evidence there too. The New York Times' Mike Isaac reported that "Google and Meta are enjoying a digital ad boom, as artificial intelligence automates marketing and drives record sales." Both companies are turning generative AI features into the connective tissue of the largest ad systems in the world — automating creative, bidding, and audience selection — and the topline says it's working.

3. Google signs the Pentagon — over a 600-employee dissent letter

The New York Times broke the news that Google has struck an AI deal with the Pentagon, despite a public letter from more than 600 Google DeepMind, Cloud, and other employees urging CEO Sundar Pichai not to provide the company's AI tools for classified military operations. Euronews reported the deal "comes as the tech giant negotiates with the U.S. Department of Defense over the potential use of its Gemini AI model in classified settings." Google spokesperson Kate Dreyer told NBC News, "We are proud to be part of a broad consortium of leading AI labs and technology and cloud companies providing AI services and infrastructure in support of national security."

The internal pushback is as significant as the deal. CBS News reported that Google employees called the classified-AI work "inhumane" and warned about uses that "would cause significant harm." This is the most concrete moment yet in which a frontier lab's commercial strategy has openly diverged from its internal AI ethics culture — and it lands the same week Anthropic's Mythos is being privately briefed to Treasury, the Fed, and the White House, and Bundesbank publicly asked the EU to seek access to it for European banks.

4. The EU AI Act stalls — Brussels can't agree on its own watered-down rules

EU member states and European Parliament lawmakers walked out of 12 hours of negotiations on April 28 with no deal on changes to the AI Act, Reuters reported. The talks are part of the European Commission's "Digital Omnibus," a package designed to simplify a stack of digital rules to "help businesses catch up with U.S. and Asian rivals." Talks resume in roughly two weeks. "It was not possible to reach an agreement with the European Parliament," a Cypriot official said.

Dutch lawmaker Kim van Sparrentak put the political reality bluntly: "Big Tech is probably popping champagne. While European companies that care about safety and did their homework now face regulatory chaos."

The sticking point — beyond philosophy — is sectoral carve-outs. Several countries and parliamentarians want industries already covered by product safety, financial, and medical regulation exempted from the AI Act's high-risk obligations on biometric ID, utilities, healthcare, creditworthiness, and law enforcement uses. Critics including privacy and civil-rights groups warn that the Omnibus, which also touches GDPR, e-Privacy, and the Data Act, is shaping up as a quiet retreat under industry pressure.

Across the Atlantic, the U.S. picture is the inverse: a federal bill is moving while states are being told to back off. CNBC reported that Rep. Ted Lieu (D-CA), who co-led the bipartisan House AI task force with Rep. Jay Obernolte (R-CA), introduced a wide-ranging bill that would tighten penalties for distributing non-consensual deepfakes, protect whistleblowers reporting AI safety risks, require U.S. participation in international AI standards bodies, and create a federal AI research prize. "It is not designed to be controversial," Lieu told CNBC. "It is based on bipartisan legislation that other members have introduced, as well as the recommendations of the bipartisan House AI Task Force."

5. The Musk vs. Altman trial moves into cross-examination — and OpenAI's IPO clock is ticking

Day 3 of the OpenAI trial saw OpenAI's lawyer William Savitt grill Elon Musk on his alleged commitment to a non-profit OpenAI and his own discussions about converting it into a for-profit subsidiary. Reuters reported Musk was due back today for a second day of cross-examination. The New York Times' liveblog described OpenAI's strategy as "targeting Musk's trustworthiness," while Musk on day 2 accused OpenAI executives of "stealing a charity."

The trial would be high-stakes in any case, but the timing is brutal: the Wall Street Journal reported on April 27 that OpenAI has "fallen short of its goals for new users and revenue in recent months," sparking concern internally about whether it can support its data-center spending. Reuters confirmed the WSJ reporting, noting it triggered a one-day selloff in Oracle and CoreWeave, who underwrite parts of OpenAI's compute. Robinhood's venture fund disclosed a $75M OpenAI investment last week, and SoftBank is reportedly seeking a $10B margin loan backed by its OpenAI stake — a sign that OpenAI's pre-IPO capital stack is increasingly leveraged and increasingly dependent on a successful public offering.

Meanwhile OpenAI shipped on the product side. The company's news index now shows "OpenAI models, Codex, and Managed Agents come to AWS" (April 28) and "The next phase of the Microsoft OpenAI partnership" (April 27) — confirming Codex distribution on Bedrock and the formal end of Microsoft's exclusivity and the AGI clause. OpenAI has also been forced into a darker spotlight: families of Canadian mass-shooting victims sued OpenAI and Sam Altman in U.S. court, alleging the company knew of the shooter's troubling ChatGPT interactions eight months in advance. Altman publicly apologized last week.

6. Diamandis on Musk: retirement savings, "universal high income," and the abundance pitch

On the human-impact track, Peter Diamandis spent the week explaining what Elon Musk actually means when he says people shouldn't bother saving for retirement. On Diamandis's own "Moonshots" podcast, Musk had said, "Don't worry about squirreling money away for retirement in 10 or 20 years. If any of the things that we've said are true, saving for retirement [will be irrelevant]." Diamandis, talking to Business Insider, reframed it as a "universal high income" thesis: AI, robotics, and energy advances drive unit costs to near-zero while corporate profits and GDP fund "COVID-style checks" to citizens.

"It's a world in which every man, woman, and child on the planet can have access to all the food, water, energy, healthcare, and education that they need and desire," Diamandis said. He sketched a near-term scenario of "$3,000 a month in universal basic income" stretched further by collapsing prices, while acknowledging a comment like Musk's "rings hollow" to people facing real cost-of-living and job-loss pressure today. "I like to say, Elon's actually never been wrong. He's just not always correct on timeframes."

The same week, Jason Calacanis kept making the more concrete version of the same argument on All-In and This Week in AI: that the cost of running an AI worker is rapidly approaching, and in some functions surpassing, the marginal cost of a human employee. "When do tokens outpace the salary of the employee? You're about to hit it. I'm about to hit it," he said in a clip widely re-circulated yesterday. MIT News added a research counterpoint with new work on debiasing vision models, privacy-preserving on-device training, and a "faster way to estimate AI power consumption" — all reminders that the technical agenda for trustworthy enterprise AI is broader than the labs' marketing.