Good morning, AI builders. ChatGPT just went from 1 billion to 1.2 billion weekly users in about 60 days, according to OpenAI's consumer product lead. He also put its search-like volume at roughly a fifth of Google's.
If you build anything that depends on search traffic, that second number is the one to pressure-test.
Welcome to the 14,154 builders reading today. Forwarded this? Join here.
The 30-second brief:
ChatGPT's numbers: 1.2B weekly users, and search near 20% of Google's by OpenAI's count
AMD's guide: about $17B next quarter as data center revenue nearly doubles
Siri AI opens up: any developer can now test app integrations in the 27.3 betas
Build with AI: stop resending your whole conversation to Claude on every turn
OPENAI
Source: Adam Fry of OpenAI, via Big Technology. Company-stated figures.
The scoop: At an OpenAI dinner in New York, ChatGPT consumer lead Adam Fry told Big Technology's Alex Kantrowitz that ChatGPT has 1.2 billion weekly active users, adding about 100 million every 30 days.
Key details:
Search: Fry put search-like query volume at roughly one-fifth of Google's, counting each question that triggers a web search.
The comparison: Kantrowitz notes Bing's search share is around 5%. Fry didn't address that directly.
Beyond chat: 250M weekly users of voice and dictation, 1.5B images a week, and ads past a $1B annualized run rate within 200 days.
The catch: these came from a brief, informal chat. OpenAI's figures aren't independently verified, and "a search" here isn't necessarily what Google would count.
Our take: Treat the 20% as a claim, not a market share, because a question that triggers a web search and a Google query are different units. But even at a discount, it means your distribution plan can't assume Google is the only front door anymore, so check how often your product shows up in ChatGPT answers.
AMD
Source: AMD Q3 2026 results. Q4 is the guidance midpoint.
The scoop: AMD reported record Q3 revenue of $13.4B, up 45% from a year earlier, and guided Q4 to about $17B, plus or minus 3%.
Key details:
Data center: a record $8.5B, up 97%, on EPYC demand and the launch of Instinct MI450 GPUs.
Customers: Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud are running workloads on the first Helios rack systems.
The weak spot: gaming fell 40% to $773M.
Our take: The customer list matters more than the beat. Having Anthropic and OpenAI both on Helios gives buyers a more credible second source to Nvidia, and the jump to $17B is the test of whether MI450 can ramp as fast as AMD is promising.
BUILD WITH AI
Stop resending your whole conversation to Claude
What you'll make: a two-turn Python chat using Message Threads, the new Claude API beta that keeps the conversation on the server and sends only what's new. Anthropic says time to first token stays lower and steadier as chats grow; token billing doesn't change.
Step by step:
Run
pip install -U anthropicand set yourANTHROPIC_API_KEY.Start the thread: call
client.beta.messages.createwith your model,system,messages,thread={"type": "create"}andbetas=["message-threads-2026-08-12"].Continue it: send only the new message, plus
thread={"type": "continue", "previous_message_id": first.id}and the same beta. Leave outsystemandtools, or you'll get a 400.Handle expiry: threads expire 5 minutes after the last request. On a 404
thread_not_found, replay the full conversation in a new create request, so keep your own copy.Ship it: log time to first token with and without threads on your longest sessions, then roll out where it helps.
Pro move: Check compliance first. Threads aren't eligible for zero data retention or HIPAA workflows, and they can't be edited or branched, so start a new thread for those cases.
APPLE
The scoop: Apple's developer preview lets any app plug into Siri AI through App Intents, starting with the 27.3 betas.
What apps get: personal context, onscreen awareness and natural-language requests, plus Skills that chain several actions together.
Timing: customers get it "in a few months" and App Store submissions open in the spring. It's unavailable in the EU on iOS, iPadOS and watchOS for now.
Our take: If your app already has App Intents, you've got a cheap head start on a new distribution channel, so test it now while the terms are still being written.
FROM THE CHANNEL
Catch up on the GPT-6.2 leaks and Gemini 4 Carbon
Watch for a fast rundown of the GPT-6.2 and Fable 5.5 chatter, which is still leaks rather than launches, plus a first look at Carbon.
⚡ Rapid fire
🏢 Salesforce agreed to buy Cresta, the voice-agent maker, for a reported $2.7B; Salesforce's own post doesn't state a price.
💰 Hark raised $700M with Nvidia, AMD Ventures and Intel Capital to build AI hardware and assistant models.
🔬 Anthropic published FlashProve, an open-source prototype that proves which model produced a response, at about 70-95x the GPU time.
🏛️ Russia ordered private data centers to reserve 10% of capacity for state use, with access required within 24 hours of an order.
👀 Unconfirmed: Nvidia told cloud clients to limit China business, reportedly requiring at least 51% of services on new contracts to go to non-Chinese firms, per the FT.
🧰 Tool drop
Copilot connectors: link Slack, Figma, Sentry and more to GitHub Copilot with one sign-in (public preview).
NorthCUA: an open-source computer-use agent family, up to 397B parameters, whose report claims wins over Qwen3.8-Max and Kimi-K3.
Google Cloud Developer plugin: gives Antigravity agents gcloud guidance and live Google Cloud docs lookups.
What Do You Rate Today's Newsletter?
That's the brief. See you tomorrow at 6am.
The World of AI team