SpaceX acquires Cursor for $60 billion, the largest VC startup deal ever

Four days after listing on the Nasdaq, SpaceX exercised an April option and agreed to buy Anysphere, the company behind AI coding tool Cursor, for $60 billion in all-stock. The deal was confirmed via SEC filing on June 16. Axios called it the largest acquisition of a venture-backed startup ever, outside of Musk’s own xAI self-deal. SpaceX stock climbed on the announcement, briefly topping Amazon by market cap at its intraday high before settling back.

Cursor’s annualized revenue grew from roughly $100 million in early 2025 to more than $4 billion by June 2026, with the company claiming over one million paying users. Because SpaceX paid entirely in stock, and that stock had gained more than 50% from its $135 IPO price in less than four trading days, the $60 billion represented less than a tenth of the market cap SpaceX had added since listing.

SpaceX absorbed xAI in February 2026, folding Grok, the Colossus supercomputer, and X into a single entity. Cursor adds $4 billion in developer revenue and access to the engineers building AI-native software. The deal raises a direct question for enterprise teams running Cursor today: the tool was built on Claude and GPT, both made by competitors of its new parent.

The full sequence behind the Fable 5 ban: a flagged investor, a refused ultimatum, a global shutoff

WIRED and The Washington Post reported the complete sequence on June 17. It ran in two steps. First, the White House flagged SK Telecom, South Korea’s largest wireless carrier and a $100 million Anthropic investor with access to Mythos 5, as a company suspected of ties to China. The administration asked Anthropic to revoke only SK Telecom’s access. Anthropic did so immediately. SK Telecom has denied any China ties and says it does not use Huawei or ZTE equipment in its core networks.

In the same week, Amazon researchers separately identified potential vulnerabilities in Fable 5, the publicly available version of Mythos, and reported them to the White House. The administration then reportedly issued Anthropic’s CEO an ultimatum to pause the model. The CEO refused. The export control directive followed, requiring a global shutoff: any model still live for domestic users remained accessible to foreign nationals working inside the US, so a worldwide takedown was the only technically compliant path.

Fable 5 remained offline as of June 20, eight days after the ban. The episode has already produced one concrete shift: enterprise risk teams are now formally treating model availability as a procurement variable rather than a constant. A US-hosted closed-weight model can be pulled globally within hours of a government directive. A model whose weights are already on servers in Frankfurt or Seoul cannot.

AI takes the G7 table: Amodei, Altman and Hassabis call for a US-led global standard

On June 17, the G7 summit in Evian-les-Bains, France held a working lunch with around a dozen AI executives seated alongside heads of state, including President Trump. Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis called for a US-led coalition of democratic countries to establish shared standards for developing, evaluating, and governing advanced AI. OpenAI CEO Sam Altman backed the proposal, calling for “an international forum for discussion that establishes globally accepted standards for testing.”

On the sidelines, G7 nations discussed a “trusted partners” access scheme that would give allied countries a pathway to Fable 5 and Mythos 5 while keeping restrictions in place for others. Those conversations ran primarily through US Commerce Secretary Howard Lutnick. Canadian Prime Minister Mark Carney indicated support for a US-led framework. No formal agreement was announced. Trump told reporters the negotiations were “going fine.”

AI executives sat at the same table as the leaders of the world’s largest economies, a shift multiple commentators described as a signal of where power in technology now sits. The Atlantic Council noted the Fable 5 ban had upended the assumption, held by multiple G7 allies, that US AI access was a baseline of the alliance. AI governance has moved out of technology policy and into foreign policy.

ChatGPT falls below 50% market share for the first time

Sensor Tower’s State of AI 2026 report, published June 16, found that ChatGPT’s share of the global AI assistant market fell to 46.4% by the end of May, the first time it has gone below 50% since OpenAI created the modern AI chatbot category in November 2022. ChatGPT held more than half the market through January 2026, and crossed below 50% in March. Gemini ended May at 27.7% and Claude at 10.3%. All other assistants, including Grok, Perplexity, DeepSeek, and Meta AI, each held less than 5%.

ChatGPT still has 1.1 billion monthly users, the fastest any app has reached that milestone. The market share data does not mean decline in absolute terms; it means a market growing faster than its leader. That framing matters for OpenAI’s S-1, filed with the SEC on June 8, which will ask investors to hold both facts at once. Gemini’s rise is driven by Android OS-level integration, not model improvements. Claude leads the field on subscriber conversion at 13%. For enterprise procurement, the practical read is simple: ChatGPT as the automatic default is behind us.

Transformer co-author Noam Shazeer leaves Google for OpenAI

On June 18, Noam Shazeer announced he was leaving Google to join OpenAI. Shazeer co-authored the 2017 paper “Attention Is All You Need,” which introduced the transformer architecture underlying every major large language model today, including GPT, Gemini, Claude, and Grok. He was serving as vice president of engineering at Google and co-lead of the Gemini model program. Google had paid approximately $2.7 billion in 2024 to bring him back from Character.AI, the startup he co-founded after Google declined to release his earlier chatbot project. He stayed less than two years.

Sam Altman said publicly that Shazeer was “one of the people I have most wanted to work with since the very beginning of OpenAI.” Per reporting by CityBiz, Shazeer will lead architecture research at OpenAI, focused on the fundamental design of next-generation models. Google said it was “grateful for Noam’s meaningful contributions.”

Shazeer is one of a very small number of people whose individual research decisions shaped how all modern AI systems are built. His move to an IPO-bound OpenAI is a direct setback for a Gemini roadmap already under competitive pressure. It also demonstrates a limit of acqui-hires as a retention mechanism: $2.7 billion did not hold him for two years.

GPT-5.4 runs a drug discovery project from lab to confirmed result. A first.

OpenAI published results on June 17 from a three-month collaboration with Polish chemistry startup Molecule.one. Working with Molecule.one’s Maria AI and an automated wet lab, GPT-5.4 independently selected a research area, generated and ranked proposals, designed experiments, interpreted results, and proposed follow-up studies for a specific medicinal chemistry problem. Human chemists steered, reviewed, and validated throughout, but did not drive the research. OpenAI and Molecule.one describe this as the first documented instance of a frontier AI model improving an open-ended organic chemistry problem end-to-end, from literature review through experimental confirmation.

The reaction in question was Chan-Lam coupling of primary sulfonamides, a copper-catalyzed method used to build carbon-nitrogen bonds found in more than 91 FDA-approved drugs. The reaction has historically returned low yields, making it a known bottleneck in early drug discovery. GPT-5.4 proposed that TEMPO, a mild oxidant, could improve results. Molecule.one’s lab ran 10,080 reactions. Average estimated yield rose from 16.6% to 25.2%, and the share of reactions clearing 30% yield rose from 15.6% to 37.5%. Human chemists validated the result across 14 substrate pairs and saw higher yields in 11 of them.

OpenAI calls the workflow near-autonomous. The system ran within the company’s Preparedness Framework, human chemists kept control of the lab, and the project was scoped to a legitimate chemistry problem. It did not produce a new drug. What it showed was a frontier model participating meaningfully in a real scientific research loop, not generating text about one. For pharma and biotech, this shifts the question from whether AI can assist researchers to how much of the loop it can now own.

Both labs build enterprise distribution networks as OpenAI files its S-1

OpenAI submitted a confidential S-1 registration statement with the SEC on June 8, the first formal step toward a public offering. No listing date has been disclosed. On June 14, the company launched its first formal global partner program, backed by $150 million and targeting 300,000 certified consultants by December 31, 2026, focused on Codex and new products from Build 2026.

Anthropic, which closed a $65 billion funding round in May at a reported $965 billion valuation, signed TCS and DXC as Global Premier partners in its Claude Partner Network on June 11, the same day. The network itself launched in March 2026 with a $100 million commitment. TCS will equip 50,000 of its associates with Claude and stand up a dedicated business unit for regulated industries. DXC will embed Claude inside the mission-critical systems it operates for banks, airlines, insurers, and governments, training tens of thousands of forward-deployed Claude-certified engineers.

Both labs are spending heavily to lock in the consulting partners who will decide which AI gets deployed inside the world’s largest organizations. Enterprise contracts in banking, healthcare, and government are where durable AI monetization will be decided. Consumer subscriptions built the user base. System integrator partnerships are how that base converts to margin.