Iran installs toll gates on the Strait of Hormuz, and OpenAI chooses to buy a media outlet instead of training a better model. When control is worth more than performance, the cost accounting of rewriting order is starting to surface.
1｜Iran puts a price tag on global oil shipping routes amid war
For the first time, U.S. forces bombed Iran’s civilian infrastructure on Thursday, destroying the B-1 bridge near Tehran. In an address the night before, Trump threatened to “blast Iran back to the Stone Age.” Almost simultaneously, Iran’s deputy foreign minister announced that it is drawing up a transit agreement with Oman for the Strait of Hormuz, charging for ships passing through. Wall Street read the move as a sign of easing, and U.S. stocks recovered all their losses. But crude oil didn’t play along: WTI rose 9% to $108.95 per barrel.
Since the war began in February, the strait’s average daily passing vessels have fallen from 130 ships to 6 ships. The IEA said this is “the largest supply disruption in the history of the global oil market.” Iran’s toll plan isn’t about easing—it turns passage rights for 20% of global oil supply from “freedom of navigation” into “conditional permission.” Diplomats from more than 40 countries are discussing reopening the route, and the U.S. refuses to attend. Trump’s stance is clear: protecting the Strait of Hormuz is not a U.S. matter.
(Source: Axios / Fortune / Al Jazeera / IEA / UNCTAD)
2｜OpenAI buys a media company—what it’s consolidating isn’t content, but narrative control
OpenAI acquired tech talk show TBPN, the first media-and-M&A case in the AI industry. TBPN streams three hours daily, positioning itself against Bloomberg and CNBC. Revenue is expected to reach $60 million this year. Sam Altman, Zuckerberg, and Peter Thiel have all appeared on the show. OpenAI claims TBPN maintains editorial independence, but the reporting party is the company’s Chief Political Operations Officer, Chris Lehane.
The logic isn’t in content, it’s in distribution. OpenAI is facing a triple squeeze: a public trust crisis, controversies over employee churn, and regulatory pressure. When the official line is “our standard PR playbook doesn’t apply to us,” the answer is to bypass the media and directly own the communication channel. Bloomberg, WSJ, Wired, Reuters, CNBC, and The New York Times all reported the deal the same day—by itself, that’s a signal: when an AI company acquires a media outlet, it’s already a matter everyone is watching.
(Source: TechCrunch / Wired / Bloomberg / WSJ / Reuters / NYT)
3｜Microsoft officially kicks off “de-OpenAI-ization,” and Suleyman pivots to superintelligence
Microsoft has completed a key reorganization of its AI team. Mustafa Suleyman is no longer in charge of the Copilot product line and is fully投入 into superintelligence model R&D. The adjustments in mid-March pushed three things at once. Copilot will be consolidated under former Snap executive Jacob Andreou. The MAI superintelligence team released three in-house foundation models: voice transcription, audio generation, and image generation. FT reports that Microsoft has also introduced “mid-tier” AI models aimed at scenarios with limited computing power.
Looking back to October 2025, Microsoft and OpenAI revised their agreement—both sides can “pursue AGI independently.” Half a year later, in an internal memo, Suleyman wrote that it should “put all its efforts into superintelligence.” This is no longer hedging risk—it’s a $1.3 billion investment, with the contract terms for an independent path now formalized.
(Source: The Verge / TechCrunch / FT / Bloomberg / CNBC)
4｜Blue Owl launches a $5.4 billion redemption wave; AI’s impact on credit assets begins to transmit
Two private credit funds under Blue Owl Capital are facing a wave of redemption requests. For Blue Owl Credit Income Corp., quarterly redemption requests jumped from 5.2% to 21.9%. Technology Income Corp. went even further to 40.7%. The company was forced to implement a 5% redemption cap, locking up about $5.4 billion. Blue Owl’s share price plunged 7%, and Blackstone, Apollo, and KKR all fell by about 3% in tandem—dragging the entire alternative asset management segment into panic.
Blue Owl blames this on “market negative sentiment,” but investors’ panic has a specific target. Blue Owl’s core asset is loans to software companies, and AI is redefining which software companies can survive and which will be replaced. When AI programming tools begin to eat into the moat of SaaS firms, cash-flow expectations behind those loans become uncertain. The $1.8 trillion private credit market has long been viewed as a “more stable fixed-income substitute.” Blue Owl is the first signal of a crack appearing in that narrative.
(Source: Bloomberg / FT / Reuters / CNBC)
5｜SpaceX kickstarts an IPO at over $200 billion; Musk ties rockets, AI, and defense together
SpaceX raised its target valuation for the IPO to above $200 billion, potentially making it the largest public listing transaction in history. Goldman Sachs, JPMorgan, Bank of America, and Morgan Stanley all received underwriting roles. The financing size could reach $75 billion—exceeding the record set by Saudi Aramco in 2019. SpaceX has filed confidential IPO documents with the SEC and expects to list this summer.
A few months ago, SpaceX’s valuation was $125 billion, and after acquiring xAI it jumped to $2 trillion. The source of the premium isn’t the number of launches—it’s that Musk has bound Starlink satellite internet, military launch contracts, and xAI’s Grok model into an inseparable integrated whole. Investors aren’t buying a space company’s stock—they’re buying entry into “Musk infrastructure.” On the same day, Tesla’s Q1 delivery numbers were released: up 6% year over year but down sequentially, and the stock fell by more than 5%.
(Source: Bloomberg / 36Kr / The Verge / SEC Filing)
Also worth noting ↓
Mercor suffers a supply-chain attack, breaking the AI training data security chain. This AI data startup, valued at $10 billion, was invaded by Lapsus$ via the open-source library LiteLLM. It claimed to have stolen 4TB of data, including training datasets and project information provided for OpenAI, Anthropic, and Meta. The attack vector was the supply chain rather than a direct intrusion, meaning that compromising an open-source component could ripple across the entire AI training pipeline. (Source: Fortune)
Google Gemma 4 switches to the Apache 2.0 license, removing one barrier to the commercial use of open-source AI. In four sizes, it is the first to adopt an OSI-approved standard open-source license, eliminating resistance from enterprise legal reviews. Previously, Gemma’s custom license pushed many enterprises toward Mistral and Alibaba’s Qwen. License changes may reshape the market landscape more than changes in model performance. (Source: Ars Technica / VentureBeat / Google)
On the same day, Trump fired Attorney General Bondi and Army Chief of Staff General George. Bondi was removed for mishandling the Epstein case and failing to prosecute political enemies; Todd Blanche served as acting attorney general. During the Iran war, Hegseth immediately removed the 41st Army Chief of Staff; George was told to “retire immediately.” Wartime leadership swaps plus political purges—two personnel lines were pulled on the same day. (Source: WSJ / Axios / CNN / NBC)
Anthropic released a research paper saying Claude has “functional emotional” representations internally. Researchers found computational structures inside the model that execute functions similar to human feelings. This doesn’t mean the AI “has emotions”—it means the model generates internal states equivalent to emotional functions when processing information. The boundary of AI safety research is expanding from “can it lie?” to “does it have some kind of internal experience?” (Source: Wired)
The CFTC sues three states to defend federal jurisdiction over prediction markets. Arizona, Connecticut, and Illinois were sued by the federal derivatives regulator on the grounds that the states have no authority to regulate federally registered markets. At the same time, the suspicious trading patterns on Polymarket that appeared within 24 hours before the war with Iran are still unresolved: six accounts bet ahead of the airstrikes and profited $1.2 million. (Source: CoinDesk / Fortune / Bankless)
Goldman Sachs quantified the labor bottleneck for AI infrastructure, and a gap of 500,000 jobs has emerged. Building data centers requires 300,000 electricity-power jobs and 200,000 grid jobs, but the U.S. currently has only 45,000 energy apprentices. Annual pay for electricians in Northern Virginia is already above $120,000. Ford CEO Jim Farley said that while AI is eliminating entry-level white-collar jobs, it is also creating a structural shortage of blue-collar labor. (Source: Fortune / Goldman Sachs)
Click to learn about job openings at Lydong BlockBeats
Welcome to join the official community of Lydong BlockBeats:
Telegram subscription group: https://t.me/theblockbeats
Telegram discussion group: https://t.me/BlockBeats_App
Twitter official account: https://twitter.com/BlockBeatsAsia
