Anthropic files for its IPO, Google issues its first new stock in two decades, and semiconductors post their best quarter ever — and right on cue, Bernie Sanders proposes the public take a 50% stake in the frontier labs. Plus: Nvidia's RTX Spark goes after the Mac, Meta's AI support bot gets Instagram accounts hijacked, and Bain finds the AI cost savings aren't showing up.
If AI is built on everyone's knowledge, should everyone own it?
Anthropic filed for its IPO the same week Bernie Sanders proposed the government take 50% of the frontier labs. With the financial stakes going vertical — Google raising $80 billion in fresh equity, semiconductors on their strongest run ever — the policy discourse is converging on Sanders's framing: not whether AI will change the world, but who will own and control that future. Even the labs have floated public wealth funds. The specifics so far are shaky, but the question isn't going away — and the discussion is going to get weirder before it lands.
Anthropic confidentially filed its IPO paperwork with the SEC on Monday. SpaceX — the only real comp at this scale — is listing about ten weeks after filing, and Axios' Dan Primack says the move puts Anthropic "on a path to go public well before Labor Day, or at least gives it the optionality."
— Sam Altman, asked Monday whether there's a race to go public. Altman's reframe: "I think there is a race to deliver the best technology and build the best business." The financial press is painting a high-stakes race with Anthropic anyway.
Renaissance Capital's Matthew Kennedy says whoever goes first sets the tone and "whoever goes second could look like an also-ran." PitchBook's Harrison Rohlfs flips it: Anthropic just volunteered to absorb all the disclosure risk, and OpenAI now has a free option to watch how institutional investors react to audited frontier-AI financials before committing to a price.
Just as every token these companies make available gets bought, every share they offer will be bought in just as short order — the market is not going to pick a winner, and both will see staggering demand. Wedbush's Dan Ives reads it the same way: "an opening of the floodgate for the IPO market."
Google will raise $80 billion in new stock to fund a $190 billion capex year, with a "significant increase" anticipated for 2027 — making it one of the first hyperscalers to ask investors to stomach dilution. The funding progression: cash on hand, then record corporate bonds, now equity.
Berkshire Hathaway signed up for a $10 billion allocation in the new issuance, lifting its Google holdings to $32 billion — a top-five position at roughly a tenth of the portfolio. The deal was hashed out over 24 hours and is Greg Abel's largest move since taking over from Buffett in January.
The S&P 500 is up 16% since the beginning of April, the fifth-strongest two-month stretch since the 1950s, driven almost entirely by semiconductors: the US Semiconductor Index gained 69% in its strongest quarter ever. Skeptics note semis are notoriously cyclical; others point to structural shortages across the AI supply chain.
In a New York Times opinion piece, the data-center-moratorium leader pivots: if we can't shut it down, the government should have a stake. Sanders advocates partial nationalization of the AI industry, arguing the models were built on "the creative work of millions of people" that has "essentially been stolen."
— Senator Bernie Sanders, in The New York Times. His framing of the stakes: will AI make life better for working families, or will the future be determined by a handful of billionaires with virtually no democratic input? "Since AI is built on the collective knowledge of humanity, the wealth it generates must benefit humanity."
A one-time 50% tax on OpenAI, Anthropic, xAI, and other labs — paid not in profits but in stock — would fund a sovereign wealth fund with voting shares and 50% control of the boards. Modeled on the Alaska Oil Fund, it would start with direct dividend payments to citizens, then fund healthcare, education, and housing.
This isn't art-of-the-deal anchoring; 50% is the number he thinks is fair. Government expropriation of important private companies might have been anathema in previous eras, but with discontent where it is, all bets are off on what the public will tolerate — AI as a public good could find real political resonance.
OpenAI's April white paper on industrial policy called for a "public wealth fund that provides every citizen with a stake in AI-driven economic growth." Anthropic wrote last October that sovereign wealth funds "could enable states to acquire positions in AI-related assets" and distribute AI-derived wealth more equitably.
— Ezra Klein, in a recent op-ed. Klein's more politically palatable strand: distribute AI itself as a public good, not just the financial upside. A public agenda for AI means defining the public problems AI can solve and creating the data, financing, and compute to solve them — "it starts with access, but it does not end there."
None of the proposals so far are convincing on the details, but the trajectory of Klein's questions — what positive benefits do we want from AI, and how do we best achieve them — is the right one. Even Bernie's maximal proposal is generating counter-responses that keep broad financial upside without putting the government in charge of the most important companies in the world. We are in uncharted territory.
Nvidia's first standalone prosumer CPU pairs twenty CPU cores with over six thousand integrated GPU cores and up to 128GB of unified memory, delivering one petaflop of AI compute (an H100 does about four). It lands in Windows PCs and laptops from Asus, Dell, HP, Lenovo, and Microsoft this fall — premium devices aimed squarely at Apple's M5-series.
— Kara Brisky, Nvidia VP for Gen AI software, on GPU-powered chatbots. The CPU is having a resurgence: GPUs are increasingly seen as hardware for training, while powerful CPUs are better at executing agentic tool calls.
OpenAI and Anthropic have already taken delivery of their first units ahead of full data-center build-outs. For the first time on an Nvidia data-center chip, the focus is the CPU: Huang calls Vera "the first CPU designed for that future, built to run agentic AI at hyperscale."
The Verge argues the RTX Spark could be Windows' M1 moment after five generations of Apple dominance in local inference. Satya Nadella used the announcement to kick off Build week, promising "unmetered intelligence to every home and every desk with Windows."
The Information reports Meta will begin testing an AI pendant over the next year as part of a "wearables for work" push — devices as the hook for Meta's AI models and consumer agent subscriptions. It's also a revenue story for Reality Labs, which lost $4 billion last quarter on $402 million of revenue.
Attackers simply asked Meta's AI support to link hijacked accounts to new email addresses, passed liveness checks with AI-generated videos, and bypassed two-factor entirely — claiming accounts including the Obama White House and Sephora. Victims couldn't reach a human tech support worker.
— Jeffrey Emanuel, on the Instagram account-hijacking exploit. Reports say Instagram's trust-and-safety org lost 60% of its people to layoffs and reassignment while "AI maxing pushed a bunch of bugs to production." You get what you incentivize — a warning for any company wanting to copy Meta.
In an April survey of large companies, almost 40% reported measured AI cost savings below 10%, despite targeting 11–20%. The blockers: 41% cite data access or integration issues, with more than a quarter flagging compliance, competing business priorities, and skills gaps.
— Bain & Company, in its April survey of large enterprises. 44% of companies are cash-flowing the next leg of AI investment on the basis of assumed cost savings — so when the savings don't materialize, the problems compound downstream.
Surging employee demand for Code Puppy — Walmart's co-worker-style agent for everything from presentations to spreadsheets — pushed the company to end its unlimited token policy and impose budgets. Walmart says it still wants employees using AI, just more efficiently; expect a lot more of this as agentic use cases meet the token-shortage era.