AI Weekly

This was one of the wildest weeks the AI industry has ever had, and I am seeing the headline numbers continually shifting. SpaceX went public in the trillions, but now governments showed they can shut down an AI model without notice, and that changes the risk for every business using AI infrastructure. Let’s see the headlines…

The Government Proved It Can Kill an AI Model

As of June 15, 2026, Claude Fable 5 and Claude Mythos 5 remain offline, three days after the US Department of Commerce issued an export control directive, with Anthropic publishing its public disagreement with the action but not announcing a resolution timeline. For the first time in history, the U.S. government forced a company to disable a publicly deployed frontier AI model, not because of a proven catastrophe, but because of a jailbreak claim, and Anthropic complied within hours. CloudflareGoogle

Every enterprise running AI in production found out their vendor risk includes a government kill switch. Finding that 16% of companies have no continuity plan if a key AI provider becomes unavailable deserves attention this week, since every enterprise using the affected models in production workflows lost access immediately and without warning. If you are building anything mission critical on a single AI provider right now, this is your wake up call to build a fallback plan. Cloudflare

SpaceX's IPO Validated the AI Infrastructure Bet

SpaceX entered week two of trading with MSCI buying active and Nasdaq-100 entry roughly ten trading days away, after debuting at a 2.1 trillion dollar day one valuation for a company that lost money last year. What a valuation tells you is where capital believes the next decade of value creation is happening. Markets are not pricing SpaceX on current profitability, they are pricing it on AI compute and infrastructure dominance. Shares closed their first trading day at $168.70 and then spiked to a high of $225 following the announcement of a 60 billion dollar Cursor acquisition, before stabilizing around $191. TechCrunchGoogle

OpenAI Killed Its Most Hyped Product to Protect Its IPO

This one is a masterclass in cutting losses before investors see the books. OpenAI announced the discontinuation of Sora, with the web and app experiences shut down and the API scheduled to close in September, despite Sora having signed a deal with Disney to license hundreds of branded characters for virtual avatars just three months before its shutdown. According to reporting, Sora burned an estimated $15 million per day in compute costs against a total lifetime revenue of $2.1 million, with active users collapsing from over one million downloads in its first week to under 500,000. CloudflareCloudflare

That is an unthinkable burn rate for a consumer product, and killing it right before a public listing tells you exactly how seriously OpenAI is managing its narrative for Wall Street. OpenAI's audited 2025 financials show 34 billion dollars in spending against 13 billion in revenue, with a net loss attributable to OpenAI of $38.53 billion. Investors are going to scrutinize every dollar, and Sora was bleeding cash with nothing to show for it. CNBC

The Subsidized Era of AI Is Over

Starting this week, Anthropic's billing overhaul separates agent and programmatic usage into metered credits, and combined with GitHub Copilot's token based switch on June 1, the message is becoming clear: the subsidized era of AI development is ending. If your business has been running lean on flat rate AI subscriptions, now is the time to actually model your usage and understand what your real compute costs look like.

The enterprise CTO question is not simple, since one developer's refactoring session could cost as much as five others' entire month of usage, which creates a structural budgeting problem for teams. I would go so far as to say this will impact every team, especially since one power user can now outpace a standard company… GoogleCloudflare

Compute Scarcity Is a Feature, Not a Bottleneck

Anthropic says it would ‘rather have more demand than it can serve than the inverse,’ and for enterprise teams modeling their AI budgets, the takeaway is that compute scarcity is a structural feature of the 2026 to 2028 period, not an anomaly as many are still hoping for. SoftBank already committed up to 75 billion euros to build five gigawatts of AI data center capacity in France, with the first phase delivering 3.1 gigawatts by 2031. TechCrunchCloudflare

This tells me anyone planning a long term AI strategy needs to stop assuming compute will get cheaper and more available next quarter. It is going to stay tight for years, and the companies that lock in supply now are the ones who will be able to scale without getting priced out later, which I believe is about everyone based on today’s rates.

What We’re Watching

Watch how the Fable 5 situation resolves, because the precedent it sets for government intervention in commercial AI is going to define policy conversations for the rest of the year. Keep an eye on how SpaceX's Nasdaq-100 inclusion plays out around July 7th since that forced index buying could move the stock significantly (read, potentiality down as most are predicting). And if your company has not stress tested what happens if your AI vendor goes dark overnight, this is the week to do it.

Stay ahead of the curve,

Clayton

Connect at claytonstrategy.com