Three new cards, one new operator. Two distinct themes from the week of June 22-26. AI infrastructure pricing is bifurcating into commoditized payment rails and genuine software businesses. Valve raised the bar for Popular Upcoming and introduced a personalized widget that changes how indie games get discovered before launch.
Theme 1, Inference pricing separates payment rails from software businesses
Two operators published on AI infrastructure this week, arriving at the same underlying question from different angles: what actually creates durable value in an AI-enabled product?
Tomasz Tunguz published "So You Want to Sell Inference" on June 22 (Reselling inference at cost-plus yields zero margin as model prices compress). The argument is direct: reselling inference at a cost-plus markup is structurally identical to being a payment processor. As model providers compete, the underlying inference cost compresses toward commodity. The cost-plus markup compresses with it. Customers who bring their own API keys eliminate the markup entirely. The only pricing model that survives is outcome-based, where the charge is a fraction of the value delivered rather than a function of tokens consumed.
"Cost-plus is a payment processor with a dashboard. Value-based is software."
Simon Willison published a case study on June 22 (An AI coding agent can convert a production ML model across frameworks and deploy it to the browser without specialist knowledge) that illustrates a related boundary shift. Directed by Willison, Claude Opus 4.8 converted a PyTorch image inpainting model to ONNX format, published it to Hugging Face, and built a WebGPU browser application, all without Willison writing the conversion code himself. The case is specific but the implication is broad: tasks that previously required ML engineering specialists are now accessible to generalists who can direct an agent through the work. The value creation in Willison's case came from direction and judgment, not low-level implementation.
"Claude Opus 4.8 is capable of converting a PyTorch model to ONNX, publishing the result to Hugging Face and then building out a web application"
Together, the two cards describe the same shift at different layers. Tunguz names the pricing consequence. Willison names the capability consequence. The common thread: infrastructure execution is becoming a commodity. Judgment about what to build and how to price the outcome is where margin lives.
- Tomasz Tunguz, Reselling inference at cost-plus yields zero margin as model prices compress. Cost-plus inference pricing compresses to zero as commoditization arrives; outcome-based pricing is the only model that preserves margin.
- Simon Willison, An AI coding agent can convert a production ML model across frameworks and deploy it to the browser without specialist knowledge. An AI coding agent directed by a non-specialist can now convert a production ML model across frameworks and deploy it to the browser, extending vibe coding into ML infrastructure work.
Theme 2, Valve's algorithm change reshuffles indie game pre-launch discovery
Chris Zukowski published "How the Steam Personal Calendar affects your launch" on June 25 (Steam's Personal Calendar widget is now the primary pre-launch discovery path for indie games without large wishlist counts). Valve raised the Popular Upcoming eligibility threshold from roughly 7,000 wishlists to roughly 100,000, effectively closing that discovery channel for the majority of indie titles. The Personal Calendar widget, which shows users a personalized upcoming release schedule based on their library and history, now serves as the replacement pre-launch discovery surface for games without large wishlist bases.
The mechanism matters for launch planning. Popular Upcoming gave smaller games a shared leaderboard to compete on. The Personal Calendar is personalized, which means reach depends on Steam's ability to match the game to the right users, not on raw wishlist count. Correct tagging and genre classification are now more consequential than they were when a high wishlist count could override weak metadata. Zukowski frames the developer community's progression through the change as "3-stages-of-algorithm-grief-acceptance" before recognizing the widget as a genuine compensating path.
- Chris Zukowski, Steam's Personal Calendar widget is now the primary pre-launch discovery path for indie games without large wishlist counts. After Valve raised the Popular Upcoming threshold to roughly 100,000 wishlists, the Personal Calendar widget became the primary pre-launch discovery path for indie games that cannot reach that count.