Custom silicon from OpenAI and Broadcom, smart glasses from Meta and EssilorLuxottica, and a GPT-5 medical breakthrough show the two labs pushing AI off the screen and into the world.
The jointly developed chip, nicknamed Jalapeno, targets the runaway cost of serving large language models at scale. It marks OpenAI's deeper move into purpose-built silicon.
The US Supreme Court ruled 6-3 that border officers no longer need clear evidence to detain returning residents, making green card status easier to strip. The decision weakens protections for long-term residents.
Built on the best-selling Ray-Ban AI glasses line, Meta introduced its glasses on June 23 with 26 styles. Meta is also planning an AI-powered prediction market app per NPR reporting.
OpenAI and Broadcom introduced a custom inference chip optimized for large language models, marking OpenAI's deeper move into purpose-built silicon. The chip targets improved efficiency and performance for LLM workloads across applications.
Continue reading arrow_outwardOpenAI and Broadcom unveiled a custom LLM inference chip, a direct bid to cut reliance on Nvidia hardware.
The Supreme Court eased deportation rules, exposing green card holders to faster removal proceedings.
Sentiment remains pinned at Extreme Fear despite a steady stream of AI product launches.
GPT-5 reportedly helped an immunologist solve a three-year research mystery, a concrete win for AI in the lab.
OpenAI and Broadcom revealed a jointly developed chip aimed at LLM inference, targeting the runaway cost of serving models at scale. The move signals OpenAI joining Google and Amazon in vertical integration, hedging against Nvidia pricing and supply constraints. Custom inference silicon could reshape margins across the entire generative AI economy.
Meta is advancing on two fronts: an AI-powered prediction market app and new Meta Glasses built with EssilorLuxottica. Both products move AI off the screen and into ambient, real-world contexts. The prediction market app in particular raises fresh regulatory questions around wagering and forecasting.
GPT-5 cracking a three-year immunology mystery and Cate Blanchett backing a human consent registry frame the dual edge of capable AI. One shows scientific upside; the other addresses the rights of people whose likeness and data feed these systems. Expect consent infrastructure to become a defining policy fight of 2026.
OpenAI's Jalapeño Cuts Costs 50%
Read full analysis arrow_outwardOpenAI and Broadcom introduced a custom inference chip optimized for large language models, marking OpenAI's deeper move into purpose-built silicon. The chip targets improved efficiency and performance for LLM workloads across applications.
Internal documents show Meta intends to release an AI-powered prediction market app, according to NPR reporting published June 25, 2026. The move would put Meta into the contested prediction market space alongside platforms like Kalshi...
In partnership with EssilorLuxottica, Meta introduced Meta Glasses on June 23, 2026, building on the technology of the best-selling Ray-Ban AI glasses line. The glasses launch with 26 styles and are compatible with prescription lenses.
Immunologist Derya Unutmaz used GPT-5 to resolve a complex research problem that had stumped him for three years, OpenAI said. The case is being promoted as evidence of AI's role in accelerating scientific discovery.
A new tool called the Human Consent Registry, co-founded by actor Cate Blanchett, launched this week to create a machine-readable record of AI permissions for likeness and voice. The registry currently has no enforcement mechanism,...
ElevenLabs released a 13-hour audiobook adaptation of Homer's Odyssey narrated by an AI voice clone resembling actor Michael Caine, using 20 other voices plus AI-generated music and sound effects. Marketing lead Jack McDermott framed...
A TDWI analysis by Gourav Singla published June 24, 2026, identifies the pilot-to-production gap as the most common reason enterprise AI initiatives stall, noting features behave differently once they meet representative user cohorts...
CrowdStrike warned that AI agent deployments carry a hidden identity problem, as autonomous agents acting with their own credentials expand the enterprise attack surface. The analysis adds to a growing focus on agent governance and...
Audit your inference spend. Map which model calls drive the bulk of your costs, then test whether a smaller or distilled model handles them. OpenAI's Broadcom chip exists because serving at scale is expensive - your code should assume the same.
Prototype an ambient AI flow. Take one task that lives behind a screen and sketch how it would work as a voice or wearable interaction, like Meta Glasses do. Off-screen UX is the next surface builders will fight over.
Benchmark a China challenger model. Run a coding task through Z.ai or Moonshot alongside your default lab to compare quality and price. They keep undercutting US labs, so know exactly where the gap is for your workload.
OpenAI is chasing efficiency in silicon while Meta chases reach in glasses and prediction markets, and GPT-5 cracking a three-year immunology mystery shows the upside is real. The same capability that solves science is why Cate Blanchett is backing a human consent registry, so the consent and trust questions will not stay theoretical. Markets stayed soft and crypto sat in extreme fear, a reminder that hardware bets and product launches do not move sentiment overnight. Builders should lower their inference costs, watch the cheaper challenger models, and start designing for AI that lives beyond the browser.
AI analysis, world news, markets, and tools. One briefing, delivered free.