Saturday, June 27
Saturday, June 27
OpenAI Used Its Own AI to Design a Chip That Makes Its Own AI Cheaper
OpenAI and Broadcom just unveiled Jalapeño, OpenAI's first custom inference chip, co-developed from initial design to tape-out in nine months, roughly half the typical ASIC timeline. OpenAI credits its own models for accelerating the process. Engineering samples are already running ML workloads at production target frequency, including a previously unannounced model called GPT-5.3-Codex-Spark. Early testing shows substantially better performance per watt than current hardware. The timing, months before a widely expected IPO, makes the strategic logic hard to miss.
OpenAI Used Its Own AI to Design a Chip That Makes Its Own AI Cheaper
OpenAI and Broadcom just unveiled Jalapeño, OpenAI's first custom inference chip, co-developed from initial design to tape-out in nine months, roughly half the typical ASIC timeline. OpenAI credits its own models for accelerating the process. Engineering samples are already running ML workloads at production target frequency, including a previously unannounced model called GPT-5.3-Codex-Spark. Early testing shows substantially better performance per watt than current hardware. The timing, months before a widely expected IPO, makes the strategic logic hard to miss.
The AI Arms Race Keeps Compressing
Happy Friday. The capital flowing into AI infrastructure has reached a scale that resists intuition, so here's some grounding:
- The combined funding announced across AI chips, talent moves, and model development in the last 72 hours exceeds the entire global venture capital total for 2015.
- Nine months from initial chip design to tape-out used to be considered physically impossible for high-performance ASICs. The old floor was eighteen months, and that was aggressive.
- Anthropic's reported annualized revenue trajectory, if accurate, represents the fastest revenue ramp in enterprise software history. Faster than Slack. Faster than Zoom during the pandemic.
- Prediction markets now price major model releases with more confidence than most earnings forecasts.
The race keeps compressing. Here's what moved this week.
Happy Friday. The capital flowing into AI infrastructure has reached a scale that resists intuition, so here's some grounding:
- The combined funding announced across AI chips, talent moves, and model development in the last 72 hours exceeds the entire global venture capital total for 2015.
- Nine months from initial chip design to tape-out used to be considered physically impossible for high-performance ASICs. The old floor was eighteen months, and that was aggressive.
- Anthropic's reported annualized revenue trajectory, if accurate, represents the fastest revenue ramp in enterprise software history. Faster than Slack. Faster than Zoom during the pandemic.
- Prediction markets now price major model releases with more confidence than most earnings forecasts.
The race keeps compressing. Here's what moved this week.
The Discourse Shaping Tech Culture Today
Security and the Open Web Under Pressure
Favorite Featured Stories
Companies with governance tooling deploy twelve times more AI projects to production. Only 4 of 13 frontier-autonomy age...
To a person with a browser, a form field is a text box. To a machine trying to operate that same page, it's a hundred li...
In 2004, a developer at ThoughtWorks wrote a JavaScript tool to stop his colleagues from manually clicking through a bil...
Playwright now ships two modes for agent-driven browser automation. One streams page snapshots into the LLM's context wi...
Companies with governance tooling deploy twelve times more AI projects to production. Only 4 of 13 frontier-autonomy age...
To a person with a browser, a form field is a text box. To a machine trying to operate that same page, it's a hundred li...
In 2004, a developer at ThoughtWorks wrote a JavaScript tool to stop his colleagues from manually clicking through a bil...
Playwright now ships two modes for agent-driven browser automation. One streams page snapshots into the LLM's context wi...