OpenAI Unveils "Jalapeno", Its First Custom AI Chip
On June 24, OpenAI announced "Jalapeno", an accelerator specialized for LLM inference co-developed with Broadcom. It is the company's first "intelligence processor", and reportedly went from design to manufacturing tape-out in just nine months.
The chip is positioned as the first step in building a compute foundation spanning multiple generations, with deployment planned to begin within the year. OpenAI bringing its inference infrastructure in-house carries major implications for reducing its dependence on NVIDIA and cutting costs over the long term. The key question going forward is how quickly the company can iterate through generations of chip design.
AI Researcher Exodus from Google Continues — Next Stop: Anthropic
Jonas Adler and Alexander Pritzel, two of Google's top AI researchers, are moving to Anthropic. This follows a string of departures by Google's leading scientists, including Noam Shazeer and John Jumper.
Google has long reigned as the world's hub of AI research, but the talent drain to rival AI companies is becoming a structural problem. Multiple factors appear to be behind the exodus, from differences in research environment to pressure toward commercialization. The flow of talent could redraw the map of research competitiveness.
Cerebras Plunges in First Post-IPO Earnings — Margin Pressure Weighs
AI chipmaker Cerebras Systems reported its first quarterly results since going public. However, its gross margin outlook for the core business came in below market expectations, stoking investor concerns and sending the stock into a sharp decline.
The CEO has pushed back, arguing that "the margin outlook is misunderstood." Still, in the AI chip market — where NVIDIA commands an overwhelming share and customers like OpenAI are moving to build their own chips — scrutiny of newcomers' profitability is likely to remain harsh.
Was "AI Will Take Engineers' Jobs" Actually True?
According to data from SignalFire, while AI-driven layoffs have been making headlines, engineers have in fact been claiming a growing share of new hires. Even as the spread of AI tools automates some tasks, overall demand for engineering appears to be holding firm.
This suggests that the narrative of "AI simply wiping out certain jobs" does not reflect reality. The nature of demand may be changing, but at least for now, engineering looks to be one of the most resilient professions in the AI era.
Google Search's AI Training Feature — Your Images Become Training Data Too
With an update to Search history, Google has introduced a mechanism that saves and uses images submitted in searches (such as reverse image searches) as training data for AI models. The feature is enabled by default.
According to the opt-out procedure reported by WIRED, users can disable this data usage from their Google Account settings. However, many users may be contributing data without ever noticing the change. Consumer awareness and consent around the collection of AI training data are being called into question once again.
"Operation Endgame": AI-Powered Takedown of Cybercrime at Scale
International law enforcement agencies and private technology companies have joined forces to carry out "Operation Endgame", a large-scale operation aimed at dismantling the cybercrime "assembly line." It simultaneously targeted two major criminal tools: the Amadey malware delivery platform and the StealC information-stealing tool.
Using AI-driven analysis, Microsoft identified that both tools depended on shared backend infrastructure, and through legal proceedings succeeded in disrupting both at once. It stands out as an example of AI proving to be a powerful tool in fighting crime as well.
---