The AI Brief

Vol. I · No. 18 · Friday, June 12, 2026

Today's brief:

  • OpenAI acquires Ona to give Codex persistent cloud infrastructure for long-running agents
  • ChatGPT's web-visit share has fallen 22 points in 15 months as Gemini surges
  • New York sends a data center moratorium and six AI-related bills to Gov. Hochul
  • SpaceX begins trading as SPCX, the largest IPO in recorded history, at a $1.75 trillion valuation
  • GitHub Copilot's token-based billing is generating 10x–50x cost shocks for developers running agentic sessions

OpenAI buys Ona to give Codex a persistent cloud workspace

Why it matters
The acquisition closes the gap between what frontier coding agents can do in a controlled demo and what they can do reliably in enterprise production: Ona's sandboxed cloud execution lets an agent keep working inside a customer's own environment after a developer's laptop is closed, addressing the single biggest blocker to autonomous multi-day agentic workflows.
What's at stake
For most operators, this is context, not a decision. For enterprises actively evaluating Codex or building agent pipelines on OpenAI's API, the Ona integration determines whether customer-controlled data boundaries and governance requirements can be met at production scale.
Decode
Sandboxed cloud execution = running an AI agent inside an isolated, pre-configured virtual environment hosted in a cloud. The sandbox contains the tools, credentials, and codebase the agent needs, keeps its work from bleeding into other systems, and stays online when the developer's local machine is off. Contrast with session-bound agents, which halt when the originating device disconnects.
Detail

OpenAI announced on June 11 that it will acquire Ona (formally Gitpod GmbH), a startup that provides secure, pre-configured cloud environments for AI agents. OpenAI is bringing Ona's secure cloud execution and orchestration technology into its Codex ecosystem, which more than 5 million people use each week to research, analyze, build, and automate their work. Financial terms were not disclosed; the deal remains subject to customary closing conditions.

Developers currently run AI coding agents on local machines, which means that when a workstation is powered off, the agents halt. That limits their ability to perform long-running tasks spanning multiple days. Ona provides a platform that enables AI agents to run in cloud-based sandboxes that remain online when developers shut down their workstations, so agents' work is not interrupted. Ona's customer-controlled execution model will allow agents to operate inside an organization's own cloud environment while OpenAI provides the intelligence and orchestration, giving organizations greater control over their infrastructure, data, and security boundaries without limiting what Codex can accomplish.

Codex now supports more than 5 million weekly active users, up from 3 million in April. The company has expanded Codex beyond software developers, adding plugins for roles including sales, investment banking, and equity investing, while knowledge workers now make up roughly one in five Codex users and are growing at more than triple the rate of the core developer group. The Ona acquisition is OpenAI's third in six months, following cybersecurity startup Promptfoo in March and health-tech company Torch in January.


54.7%
ChatGPT's share of worldwide AI chatbot web visits in April 2026

ChatGPT's commanding lead has shrunk by 22 points in 15 months

Why it matters
The erosion from 76.5% in February 2025 to 54.7% in April 2026 is the fastest large-platform market-share decline in recent tech history, and it is almost entirely attributable to Google Gemini's rise from 5.6% to 27.4%, driven by default placement across Android, Search, and Workspace rather than model quality alone.
What's at stake
The distribution advantage that Google has systematically converted into share is the same mechanism that could determine whether OpenAI's IPO investor thesis holds: a platform competing on distribution is structurally harder to displace than one competing on capability alone.
Detail

As of the June 2026 edition of Momentic's report, ChatGPT leads at 54.7% of worldwide web visits across the seven largest generative AI chatbots, ahead of Google Gemini at 27.4% and Anthropic's Claude at 8.2%. The report defines "market share" as web-visit share: each assistant's portion of measured web traffic across a defined seven-app set, as estimated by Similarweb. The seven apps drew a combined 10.07 billion web visits in April 2026.

Claude's numbers are the most striking in growth terms: 8.2% worldwide web-visit share, but growing 306% in a single quarter, from 203 million web visits in January 2026 to 824 million in April 2026. In the United States specifically, Claude's web-visit share is 12.5%, ahead of its global average. DeepSeek held about 4.1% of worldwide AI chatbot web visits in April 2026, but only 1.1% in the United States, reflecting a heavily Asian user base.

Caveat: web-visit share is a legacy metric. The metric measures each assistant's slice of traffic to its consumer website, an inheritance from the search era when popularity meant visits to a page. Web-visit share excludes apps, APIs, and embedded usage, so it understates total use , particularly for Claude (predominantly API and enterprise) and Gemini (predominantly Android and Workspace surfaces). Momentic is a marketing analytics firm with vendor relationships; its data is sourced from Similarweb, a third-party traffic measurement panel.

Momentic: June 2026 Top Generative AI Chatbots by Market Share (primary)/ TechTimes: ChatGPT Holds 54.7%, Claude Up 306%: Why That Number Misleads/ CaveatMomentic is a marketing firm with vendor relationships; Similarweb panels measure web traffic only and do not capture app, API, or embedded assistant usage.

New York passes a data center permit freeze and six AI bills, all awaiting Hochul

Why it matters
If signed, New York's one-year moratorium on permits for large-scale data centers would be the first such freeze in the nation, creating a direct conflict between the state's AI infrastructure ambitions and the legislative response to rising energy costs in communities near existing facilities.
What's at stake
For most operators, this is context, not a decision. For hyperscalers and co-location developers with New York permitting timelines in 2026, the moratorium window (if signed) could delay or redirect capital plans, while the FAIR News Act and training-data transparency bill set disclosure obligations relevant to any operator publishing AI-generated content in the state.
Detail

The New York legislature adjourned on June 5 after passing seven AI-related bills, all of which now sit with Gov. Kathy Hochul, who has until Dec. 31 to sign or veto the measures. The most consequential is the Responsible Data Center Development Act (S10642/A11560). The measure passed the Senate 44-16 and the Assembly 102-39 in the final hours of the session. It bars the Department of Environmental Conservation from issuing permits, certificates, registrations, licenses, or other approvals to large data centers for one year after the law takes effect. Renewals and modifications of existing approvals are exempt, as are projects that have already commenced construction by the effective date.

The moratorium targets facilities with a peak demand of 20 megawatts or more. The bill also establishes new labor standards for data centers with peak demand over five megawatts: construction projects must use only iron and steel manufactured in the U.S., transition to renewables by 2040, and pay prevailing wages. Hochul has expressed reservations about a statewide moratorium, suggesting decisions should be left to municipalities. She has not yet said whether she will sign or veto the bill.

The six other AI-related bills passed by the legislature include: a kids chatbot safety bill (S9051) that would prohibit AI companies from offering companion chatbots to minors, which passed unanimously 137-0 in the Assembly and 60-0 in the Senate; the AI Training Data Transparency Act, which requires generative AI developers to post a high-level summary of the datasets used in the development of their models; and the NY FAIR News Act (S.8451-B/A.8962-B), which mandates that news organizations operating in New York fully disclose to the public when they use generative AI to create news content, articles, or media. Penalties under the FAIR News Act range from $1,000 for a first violation to $5,000 for subsequent offenses, per the bill text.

NY Senate: Responsible Data Center Development Act (S10642) (primary)/ NY Senate: FAIR News Act passage (primary)/ Transparency Coalition: AI Legislative Update June 12, 2026/ NoteBills have passed both chambers but are not yet law. Hochul has until Dec. 31, 2026 to act on each measure.

SpaceX begins trading as SPCX at the largest valuation ever accorded a new listing

Why it matters
SPCX's debut at a $1.75 trillion valuation resets the benchmark against which Anthropic and OpenAI, both of which have filed confidential S-1s, will be priced by institutional investors over the coming months.
What's at stake
The comparison axis investors will apply is stark: SpaceX's Starlink division generates $11.4 billion in revenue at a 63% EBITDA margin, while the xAI division absorbed into SpaceX ran a $6.36 billion operating loss in 2025, making the AI-infrastructure portion of the SPCX bull case directly analogous to the unproven revenue growth paths Anthropic and OpenAI will have to argue in their own roadshows.
Detail

SpaceX officially hit the market on June 12 with an IPO price of $135 a share. This marks the biggest IPO ever, raising $75 billion at a post-money valuation of $1.75 trillion, which dwarfs the previous $35.4 billion record set by Saudi Aramco in 2019. Goldman Sachs led a 21-bank syndicate, with 30% of the float allocated to retail investors through Robinhood, Fidelity, and Charles Schwab. Elon Musk did not sell a single share.

SpaceX bought xAI, Musk's artificial intelligence company that owns X (formerly Twitter), in February 2026 in an all-stock deal valued at roughly $250 billion. In May, Musk announced that xAI is fully absorbed by SpaceX and will rebrand as SpaceXAI. With only a 4% float, MSCI inclusion starting June 13 will create structural demand for index funds. According to some estimates, the IPO is 3.3 times oversubscribed, meaning demand for SPCX stock more than tripled the number of shares available for purchase.

The competitive framing matters for AI operators tracking the broader IPO wave. Anthropic's confidential S-1 filing on June 1, 2026, ahead of OpenAI's June 8 submission, has strengthened trader consensus around its 78.5% implied probability of completing an IPO first, per Polymarket. OpenAI projects burning $85 billion in 2028 even after doubling sales from the year prior, per The Wall Street Journal. Put another way, OpenAI is asking public market investors to buy into a business that, by its own projections, won't generate more cash than it spends for at least four more years.


GitHub Copilot's token meter has handed agentic developers a bill shock

Why it matters
GitHub's June 1 switch from flat-rate subscriptions to token-based billing exposed a structural truth that vendors had obscured: AI coding assistance, once an agent can run for hours and spawn sub-agents, is priced like cloud compute, not like software, and the delta between those two cost curves can be an order of magnitude.
What's at stake
For most operators, this is context about where the developer tools market is heading. For engineering leaders who approved Copilot at flat-rate pricing and have since encouraged agentic workflows, this is a budget line that needs immediate review before the first post-transition invoice arrives.
Decode
Token-based billing = charging users for each unit of text (input and output) processed by the AI, rather than charging a flat monthly fee. A single agentic coding session that pulls in a large codebase as context and generates extensive output can consume thousands of tokens and cost multiples of an entire month's flat subscription. The switch makes cost proportional to actual compute consumed.
Detail

GitHub Copilot, Microsoft's AI-powered coding assistant used by millions of developers, shifted to token-based billing on June 1, replacing its flat-rate premium request system with metered "AI Credits" that charge based on actual token consumption. The change, announced April 27 by GitHub Chief Product Officer Mario Rodriguez, keeps base subscription prices unchanged but fundamentally alters how developers pay for the tool's most advanced features. The shift applies to 4.7 million paid subscribers. Developers running agentic sessions project cost increases of 10x to 50x, and the fallback model that previously kept users working on a cheaper model has been removed.

Users are charged based on how many tokens they burn through as they work. Some developers with financial whiplash have taken to Reddit and X to bemoan what, in many cases, appears to be a drastic escalation in cost. "What a joke," one Redditor wrote, claiming that while they currently pay around $29 per month, the new rate would balloon their costs to nearly $750 a month. Internal Microsoft documents obtained in April suggested the week-over-week cost of running GitHub Copilot had nearly doubled since January 2026, making the transition more urgent than a planned product strategy would suggest.

Tab-completion-heavy users, the majority of Copilot's user base by most estimates, are unlikely to see any meaningful increase. Developers who built daily routines around multi-step agentic sessions face the steepest recalibration, precisely because those are the workflows GitHub promoted most aggressively over the past two years. Cursor at $20 per month and Windsurf at $15 per month are the closest direct alternatives for IDE-first agentic workflows. Open-source tools including Continue.dev, Cline, and Aider allow developers to connect their own API keys and pay inference costs directly without a platform markup.

TechCrunch: GitHub Copilot's new token-based billing spurs consternation among devs (primary)/ TechTimes: GitHub Copilot Pricing Change Drives Backlash/ NoteCost projections from developer community reports; actual bills vary by workflow type and model choice. Tab-completion users are unlikely to see meaningful increases.