A big day in AI: OpenAI unveiled its first custom chip, GPT-5.6 got a restricted preview, Google lost four senior researchers to Anthropic in one week, and Anthropic accused Alibaba of the largest known AI model theft attempt. Here's everything that happened.
New Hardware
OpenAI breaks free from Nvidia dependence with its own silicon
OpenAI and Broadcom formally unveiled Jalapeño, OpenAI's first custom-designed AI inference chip, at a ceremony in San Francisco where Broadcom CEO Hock Tan personally delivered engineering samples to Sam Altman and Greg Brockman.
The chip is designed specifically for serving large language models — not training them — with the goal of cutting inference costs by 50% compared to current Nvidia GPU setups. It's already running workloads including an unreleased GPT-5.3-Codex-Spark model in lab testing.
Timeline: small prototype data center deployment by end of 2026, production ramp in 2027, full scale by early 2028. This is a direct challenge to Nvidia's dominance in AI compute — and signals OpenAI is serious about controlling its infrastructure costs long-term.
What it means for you: Cheaper AI chips = cheaper AI services eventually. OpenAI building its own hardware is a sign the company is planning for massive, long-term scale.
New Model
OpenAI's most advanced model yet — but you can't access it
OpenAI revealed GPT-5.6, its most advanced model to date, but with a catch: it's only available to a small group of customers approved by the Trump administration while it undergoes national security review under a new federal AI oversight framework.
The model is expected to reach broader availability within weeks once the review completes. No benchmarks have been published yet, but OpenAI describes it as a significant leap over GPT-5 in reasoning and code generation.
What it means for you: If you're a ChatGPT Plus user, expect GPT-5.6 access within the next few weeks. Worth watching — early reports suggest major improvements in coding tasks.
Talent
The exodus that wiped $270B from Alphabet's value
Two more senior Gemini researchers announced they're joining Anthropic today — making it four high-profile Google AI exits in just six days. The departures triggered a significant sell-off in Alphabet shares, with $270 billion wiped from the company's market cap over the week.
The pattern suggests Anthropic is aggressively recruiting from Google's DeepMind and Gemini teams, likely accelerating with the news that Anthropic is approaching its first operating profit. Meanwhile, Google's Gemini 3.5 release has reportedly been delayed, possibly connected to the team instability.
What it means for you: Anthropic is becoming the destination for top AI talent, which bodes well for future Claude improvements. If you haven't tried
Claude AI recently, now's a good time.
Legal
28.8 million fraudulent API calls — the largest known AI model theft attempt
Anthropic sent a formal letter to US Senate Banking Committee leaders accusing operators affiliated with Alibaba and its Qwen AI lab of conducting the largest known "distillation attack" on Anthropic to date — running 28.8 million fraudulent exchanges against Claude to extract its outputs and use them to train competing models.
The practice of distillation — using one AI's outputs to train another — is increasingly common and increasingly contested legally. Anthropic's decision to go to the Senate signals they're pushing for regulatory action, not just a legal dispute.
What it means for you: This is the clearest sign yet that AI model IP protection is becoming a major legal battleground. Expect more regulatory action around AI training data in 2026.
Business
$47B annualised revenue run rate — profitability in sight
Anthropic is tracking toward its first operating profit of approximately $559 million in Q2 2026, with an annualised revenue run rate of $47 billion. The company has grown faster than most AI observers predicted — and the talent inflow from Google only strengthens its position heading into the second half of 2026.
What it means for you: A profitable Anthropic is a stable Claude. Good news if Claude is your primary AI tool.