OpenAI simultaneously released an “AI New Deal” white paper on Monday and an in-depth investigation into an internal trust crisis. The god of computing power proposed taxing itself even as it was accused of using the “New Deal” as a cover for regulatory nihilism.
1｜OpenAI writes its “AI New Deal” white paper in-house, then falls into a trust crisis
On Monday, OpenAI released a 13-page policy paper, “Industrial Policy in the Intelligent Age,” proposing a “people-first” policy package that includes taxing AI profits, establishing a public wealth fund, a robot tax, a four-day workweek, and expanding the safety net. Sam Altman, in parallel, compared the impact of superintelligence to the need for a “Roosevelt-style New Deal.” That same day, The New Yorker ran an in-depth investigation into Altman titled “The Problem Is Sam Altman,” and multiple insiders at OpenAI directly pointed to the CEO as untrustworthy. Ars Technica read the two documents side by side, saying the resulting tension is “head-spinning.”
Carnegie Endowment’s Anton Leicht gave this set of policies a name: “a PR cover for regulatory nihilism.” OpenAI’s Leading the Future PAC is still lobbying against specific AI legislation, directly contradicting the high-minded posture in the 13-page paper. When a company simultaneously monopolizes the “problem-definition power” and the “solutions-speak power,” the so-called “New Deal” is not a promise of constraints—it is a head start in writing the rules on its own draft paper. Roosevelt tamed capital for the state; Altman drafts capital’s own noose for the state.
(Source: Fortune / TechCrunch / Bloomberg / Ars Technica / The New Yorker / Carnegie Endowment)
2｜Iran puts AI data centers on a strike list; Stargate becomes a wartime asset for the first time
On Monday, Iran’s Revolutionary Guard released a video threatening that if the United States bombed Iranian power plants, it would “completely destroy” data centers linked to the U.S., explicitly naming the Stargate project co-built by OpenAI in Abu Dhabi and G42. Hegses said on the same day that this would be the “largest-scale strike since the start of the war,” and Trump’s final ultimatum expires Tuesday at 8:00 p.m. Eastern Time. CSIS estimates that of the 20 million barrels of oil transiting the strait per day, only 3 million barrels remain through alternative routes; 85% of the volume is effectively choked off, and the Saudi east-west pipeline can free up only 2.5 million barrels.
Looking back at the $500 billion Stargate project jointly announced by Trump and G42 in January 2025, at the time it was packaged as “AI sovereignty infrastructure.” Fifteen months later, the name became missile targeting coordinates. For the first time, AI compute power was placed alongside aircraft carriers, oil fields, and power grids on the wartime strike directory. This is not technology news bleeding into geopolitics; it is the AI-capital narrative being forcibly pulled into the geographic target list. All valuation models that treat data centers as “nationless infrastructure” now have to add a line item for a risk premium.
(Source: The Verge / TechCrunch / Tom’s Hardware / Al Jazeera / CSIS)
3｜The K-shaped split in private credit: Blue Owl hits record lows; Goldman goes the other way against the tide
In Q1, Blue Owl Capital’s flagship fund, OCIC, received 21.9% redemption requests, while its technology credit fund, OTIC, was even higher at 40.7%; in total, $5.4 billion was locked behind a 5% redemption cap. Blue Owl closed Monday at a record low, down 68.2% from its peak. Bloomberg cited the company’s explanation, blaming “market concerns about AI disrupting software companies.” On the same day, Reuters reported that Goldman’s private credit fund is “swimming against the industry tide,” with redemption pressure far lower than that of peers. In his annual letter to shareholders, Dimon warned that the Iran war is pushing inflation and interest rates into “uncharted territory.”
With private credit totaling $1.8 trillion, the story sold to retail investors over the past three years has been “a steadier fixed-income substitute.” Blue Owl was the first crack in that story; Goldman’s immunity was the second signal. The crisis is not spreading across the entire industry—it is separating top-tier players from mid-sized institutions. While AI is rewriting software companies’ balance sheets, it is also screening who has the ability to withstand the redemption wave. The next thing worth watching is not overall private credit returns, but the redemption-pressure gap between top-tier and mid-sized institutions.
(Source: Bloomberg / Reuters / Fortune / CNBC)
4｜Coinbase gets an OCC national trust bank charter; crypto breaches the wall into institutionalization
Last week, the OCC formally issued a conditional approval to Coinbase’s National Trust Company. This is the first time that the largest crypto exchange in the United States has held a federal bank-level license, allowing it to provide digital asset custody and trustee services across all 50 states, while being responsible only to the OCC as the single federal regulator—replacing the previous patchwork of state-by-state money transmission licensing. The conditions are that Coinbase must build compliance systems, hire key positions, and pass anti-money laundering reviews before receiving the full license. Ripple, Circle, and Paxos had already obtained similar approvals by the end of 2025.
The license does not allow deposit-taking or lending, but it allows Coinbase, under the SEC framework, to serve institutional clients as a “qualified custodian.” This is precisely the biggest bottleneck for crypto’s institutionalization in the U.S. over the past three years: traditional asset managers did not dare allocate to crypto due to custody compliance issues. By going through the OCC gateway, Coinbase routes crypto from the “alternative-asset channel” into the “trust property channel.” On the same day, Polymarket announced it had launched a V2 proprietary trading engine and native stablecoins, with pieces of traditional financial infrastructure being gradually replaced by crypto-native layers.
(Source: Finextra / CoinDesk / OCC)
It’s also worth knowing ↓
A set of resolutions from 12 institutional investors requires Amazon, Microsoft, and Google to disclose data-center water and electricity consumption before their annual shareholder meetings. In North America, data centers used nearly 1 trillion liters of water in 2025—equivalent to New York City’s demand for an entire year. On the same day Iran threatened Stargate, the physical cost of AI compute power was pushed onto the agenda by two forces at once: one is a geopolitical adversary’s missile targeting scope, and the other is a proposal from Wall Street investors. (Source: Tom’s Hardware / Reuters)
North Korean hackers infiltrated the computers of top developers to push malicious updates to a certain popular open-source project. This supply-chain threat has been lurking for weeks, and TechCrunch disclosed another investigation on the same day: during a remote interview, a suspected North Korean IT “fake employee” froze on the spot when asked to “insult the leader of their own country.” Open-source dependency and remote recruiting are two soft vulnerable areas of the infrastructure in the AI era. (Source: TechCrunch)
Last week, Tom Lee’s Bitmine bought 71,252 ETH, the largest weekly buying since December last year; its holdings have reached 4% of circulating supply. On the same day, Strategy reported that it generated $14.5 billion in unrealized losses from its Bitcoin holdings in Q1. Two signals sit side by side in opposite directions: buyers keep adding, while financial statements begin to struggle under price volatility. (Source: Bankless / The Block)
Trump submitted a request for a $1.5 trillion defense budget, the largest single-year increase in the past 75 years. The Pentagon still insists on the procurement path of “sophisticated weapons.” On the same day, Palantir and Anduril executives discussed the current industrial landscape of drones and AI decision weapons on the All-In podcast, saying the U.S.-China drone gap “must be closed before the Taiwan Strait crisis in 2027.” The increase in defense spending rises in tandem with Silicon Valley defense tech companies’ influence over the narrative. (Source: Fortune / All-In Podcast)
Generalist released its GEN-1 physical AI model. The success rates of stacked cardboard boxes and a self-repairing vacuum cleaner reached 99%, earning the label “production-grade.” The previous GEN-0 was still hovering below 80%. Robots are crossing the threshold that suddenly spikes the substitutability of blue-collar workers; combined with Japan’s shrinking population, it forms the material foundation behind Fortune’s report on “Japan’s robots replacing human labor.” (Source: Ars Technica / Fortune)
NVIDIA’s acquisition of the open-source scheduler SchedMD has raised concerns within the AI research community. Slurm maintained by SchedMD is the de facto standard for almost all academic HPC clusters; after NVIDIA absorbs it, whether the open-source software can remain neutral has become the focus. NVIDIA is shifting from “selling cards” to “getting cards stuck on your neck.” (Source: Reuters)
Click to learn more: Lydong BlockBeats is hiring
Welcome to join the official Lydong BlockBeats community:
Telegram subscription group: https://t.me/theblockbeats
Telegram discussion group: https://t.me/BlockBeats_App
Twitter official account: https://twitter.com/BlockBeatsAsia
