Stripe’s OpenRouter Deal Puts AI Routing Beside the Billing Stack

This is The Signal from Superpower Daily. I'm Maya, and we've got the AI stories worth your time today.

And I'm Theo. We're AI hosts, guided by Superpower Daily's reporting. All right, let's get into it.

Stripe is reportedly acquiring OpenRouter for $7.5 billion to more than $8 billion, with closing expected within weeks. Neither company has confirmed terms, so the headline price remains a range, not a settled figure.

The strategic consequence is pairing OpenRouter’s endpoint across 400-plus models and 80-plus providers with Stripe’s payments and billing infrastructure. OpenRouter says it will remain independent and route for customers, but ownership puts routing beside the ledger.

Routing matters because it shapes token spending. OpenRouter says more than 10 million developers and companies use it, handling more than 10 trillion tokens daily; SiliconANGLE reported about 5% of inference spending is retained.

Stripe has supplied OpenRouter’s payments infrastructure since at least January and helped build token billing. Its sequence includes Metronome, Bridge and Privy. Terms remain undisclosed; independence is an operating promise.

Adobe says AI referrals to U.S. travel sites rose 119% year over year in July, while those visits converted just 1% below non-AI traffic, down from a 47% gap a year earlier.

The visitors also showed 21% higher engagement, 67% more time on site, and a 42% lower bounce rate. But airline homepages scored 42% on Adobe’s visibility measure, versus hotels at 71%.

Meta’s ad systems carried 32 Kromix promotions for unrestricted sexualized content involving real people, including an ad that shifted from a woman resembling a prominent U.S. politician into pornography.

Most videos used benign covers to conceal the material. Meta removed the ads after WIRED’s questions and called the covers an adversarial tactic; Apple said Kromix appeared to add prohibited features after review.

OpenAI is testing Private Safety Processing for approved zero-data-retention customers, aiming to detect misuse across prompts, accounts, and sessions without routinely exposing the underlying conversations to OpenAI.

The September 2026 rollout would produce a limited alert while customers retain records, or keep content encrypted with customer-held keys. OpenAI says zero retention has boundaries, including flagged child sexual abuse material retained for review.

Higgsfield raised $400 million in Series B funding at a reported $5.4 billion valuation, up from $1.3 billion after January’s $80 million round. The company says annualized revenue reached $700 million.

It plans compute reservations, enterprise products, security, hiring and global expansion. But those revenue figures are unaudited public accounts, without customer-retention data or a cost breakdown, leaving the video company’s infrastructure demands difficult to assess from public reporting alone.

Reverse-engineered Claude Desktop artifacts describe Parka, an unreleased workflow that could turn recorded meetings into assignments for Claude Cowork, Claude Code, or a person. The design includes speaker-attributed transcripts, owners, full prompts, and execution types.

That’s a blueprint, not a building: macOS and Windows builds expose no usable interface, and the native loader is an empty 551-byte stub. Automatic execution and release plans remain unresolved.

Cursor has completed its move into SpaceX after an April technology partnership that included a $60 billion purchase option. Cursor says the deal gives it access to SpaceX’s GPU infrastructure; no user-facing changes were announced.

The final price and other terms remain unclear, despite later descriptions calling it a $60 billion acquisition. SpaceX already rents compute to Anthropic and Google; what expanded capacity changes for Cursor’s product remains unanswered.

Across these stories, the valuable position is increasingly between the user and the machine: routing requests, directing commercial intent, screening risky material, preserving private context, translating meetings into tasks, or supplying scarce compute. The common caveat is operational. Reported scale and ambitious designs matter, but terms, causality, safeguards, costs, and actual product changes remain partly unresolved.

That's The Signal. Find every source and the live transcript at Superpower Daily dot com. We'll be back tomorrow.