Miro at $1.36B, Oracle’s $7.5B U-turn, and an Anthropic PR op

Three stories this week and one thread running through all of them: the price sheet is where the real story shows up, not the press release. A trophy SaaS sold for a tenth of its 2021 tag. A cloud vendor beat every AI-capex estimate on the tape, and its founder filed to sell $7.5 billion the next day. And a viral "AI whistleblower" resignation looks less like conscience and more like a coordinated campaign to shape who gets to build frontier models.

Bending Spoons buys Miro for $1.36B — 92% below the 2021 tag

On September 10, Bending Spoons agreed to acquire Miro for $1.36 billion in cash ($1.79 billion equity value), a 92% cut from the $17.5 billion Miro raised at in early 2022. The interesting part is what did not shrink. Miro today runs at roughly $600 million ARR (90% enterprise), 4 million paying customers, 100 million total users, $435 million net cash, and it is profitable. The business roughly quadrupled since the 2021 raise. What repriced is not the company, it is the growth story the 2021 market bought.

This is Bending Spoons' second trophy pickup this quarter. On August 4, the same buyer took Airtable for $1.28 billion, down from $11 billion in 2021. The Italian acquirer has now built a portfolio that reads like a museum of the last SaaS cycle — Evernote, WeTransfer, Meetup, StreamYard, Splice, Filmic Pro, Airtable, and now Miro.

The pricing read matters more than the deal size. Bending Spoons' operating playbook, wherever they land, is monetization. Evernote's paid tiers went up and its free plan shrank after the 2023 acquisition. Meetup's organizer subscription rose sharply. WeTransfer trimmed its free product. If you renew Miro or Airtable in the next 12 months, model a hike into the base case. The acquisition headline is not the price event — the invoice 12 months from now is.

Oracle prints the AI quarter of the decade, and Ellison files to sell $7.5B the next day

On September 10, Oracle reported Q1 FY27 revenue of $19.3 billion, up 30% year-over-year, with cloud infrastructure up 121% to $7.4 billion and $30 billion of new AI contracts booked in the quarter. Adjusted EPS came in at $1.92 versus the $1.74 consensus. The company guided full-year capex to $90–95 billion and full-year revenue above $90 billion. The stock rallied more than 4% after hours on the print.

The next day, September 11, a 10b5-1 filing disclosed that Larry Ellison had adopted a trading plan on June 22 to sell up to 50 million Oracle shares, worth roughly $7.5 billion at recent prices, with the plan running through October 24. Ellison had not sold more than 25,000 shares at a time this entire century. Twenty-four hours after the disclosure, on September 12, Oracle announced Ellison had cancelled the plan. No shares changed hands.

The pricing signal is in the capex line, not the sell plan. Oracle is committing $90–95 billion this fiscal year to build the datacenters behind those AI contracts — up from the $35 billion it projected months earlier. That is the capacity that ultimately sets Oracle Cloud Infrastructure's rate card, and it prices the ceiling for what hyperscalers can charge for GPU-hour compute. Ellison's aborted $7.5 billion cash-out is the noise. The signal is that the person best positioned to know what an OCI customer will pay in 2028 is spending on capacity, not selling stock.

The Anthropic "whistleblower" storm reads like a coordinated PR op

On September 8, a 27-year-old researcher named Jacob Coxon resigned from Anthropic and posted a seven-part X thread warning that AI companies are "gambling with our lives" and that AI "could kill us all by the end of the decade." He had been at Anthropic four months — two months short of his equity vesting. The post cleared 150 million views. Bernie Sanders and Governor JB Pritzker quote-tweeted it within a day. The Wall Street Journal ran a briefed piece on the resignation. Evan Hubinger, Anthropic's Alignment Science lead, publicly co-signed with a "greater than 10% chance AI kills all humans" figure.

The pattern reads as a coordinated campaign, not a spontaneous act of conscience. The Coxon account had no prior posts and no followers before the thread. Within fifteen minutes it was amplified by three organizations — Encode AI (the group behind California's SB 53), the AI Policy Network, and the AI Futures Project — all funded by Jaan Tallinn, who co-led Anthropic's Series A. The Wall Street Journal story was pre-briefed under embargo. Nvidia's Jensen Huang, speaking at Goldman's conference the same week, called the claims "outlandish and deeply untrue, wrong, arrogant, and ignorant."

The pricing angle is regulatory capture. The stated policy ask across this cluster — a federal AI regulator, a licensing regime for frontier models, restrictions on open-source model weights — would coalesce frontier model supply into a duopoly. That structure supports elevated pricing for Anthropic and OpenAI (and their Series A investors) and stalls the open-source deflation curve that has driven inference costs down roughly 10,000x in three years. The context makes the timing awkward: Anthropic is in an active S1 process targeting an IPO at a trillion-dollar-plus valuation while its own alignment lead publicly says the product may end civilization. That is not a normal risk-factor disclosure. Either the company disavows Coxon and the frontier-safety framing collapses, or it agrees with him and no public-market investor should be underwriting the IPO. The tension is real, and it is priced into everything the labs charge — and everything they will charge once the moat is legislated.