Good morning. It's Thursday, July 16, and we're covering China's massive AI chip IPO, Google's escalating hardware battle with NVIDIA, and Anthropic's latest push to strengthen AI safety.
Plus: we're exploring a live, hands-on AI course for teams (tell us if you're in), and the verdict on yesterday's Real or AI challenge.
Anthropic is recruiting experts in nuclear security, explosives, cybercrime, and fraud to stop its AI from enabling dangerous misuse. The company has 32 open safety roles focused on stress-testing models before release. Many positions pay in the mid- to upper-$200,000 range, while similar OpenAI roles offer $295,000–$445,000 salaries. The hiring underscores how AI labs are taking a larger role in building safeguards.
Apple Intelligence is entering China by relying on Alibaba's Qwen model rather than Apple's own AI. China's Cyberspace Administration approved the service, enabling AI features across iOS, iPadOS, macOS, and visionOS. The launch follows stalled talks with Baidu and evaluations of DeepSeek and ByteDance models. The approval comes as Apple's Greater China revenue rose 28% to $20.5 billion in the second quarter.
Thinking Machines Lab released Inkling, its first foundation model, more than a year after Mira Murati founded the company and raised a record $2B seed at a $12B valuation. Inkling is multimodal, open-weight (developers can download and customize the weights), and built from scratch on NVIDIA's latest infrastructure, positioned as a customizable model designed to balance "cost against performance," not to top state-of-the-art benchmarks against frontier peers.
SoftBank CEO Masayoshi Son argues the AI boom will require about $5 trillion in annual investment through 2040 and insists it is not a bubble. He says AI could generate 20% of global GDP by 2040, making today's spending worthwhile. Son also forecasts AI data centers will need 3 terawatts of power and predicts as many as 100 trillion AI agents will exist by 2040.
POWERED BY BOX
Turn Your Content Into Actionable Insights
Enterprise AI only works when it has the right business content. Put your unstructured data to work to unlock critical insights from content, automate tasks, accelerate decisions, and help teams work smarter — while keeping sensitive content protected every step of the way with the Box Agent.
Your business lives in your content. Learn how the Box Agent unleashes it.
FROM THE ZEITGEIST
When 9 Million Users Still Isn't the Whole Story
Codex reportedly jumped from under 6 million users to more than 9 million in about four days, and nobody in the room was ready to take the number at face value. That's where this week's Zeitgeist starts, and it only gets weirder from there:
A copy-paste prompt that hit 4,000 views per minute
How a modest production edge creates an enormous attention advantage
And a Codex workflow that edits video inside Premiere
IPOs
China’s CXMT Seeks $9.8 Billion IPO to Expand AI Chip Ambitions
China’s largest memory chip maker, ChangXin Memory Technologies (CXMT), is seeking to raise up to $9.8 billion in an initial public offering on the Shanghai Stock Exchange, positioning it for what could become Asia’s largest IPO of 2026. The company is benefiting from surging demand for AI memory chips, with first-quarter revenue soaring more than 700% year over year to nearly $7.5 billion, while posting over $1 billion in profit in 2025 after a multibillion-dollar loss just two years earlier.
Backed by significant government support, CXMT has grown its global memory chip market share from 3% to 8% in a year, though U.S.-led export restrictions continue to limit its access to advanced chipmaking equipment. The offering underscores China's push to build a self-sufficient semiconductor industry as competition intensifies worldwide. → Read the full article here. (Paywall)
DEPLOYMENT
Anthropic and Blackstone Launch Ode to Bring AI Into Enterprises
Anthropic and Blackstone have launched Ode, a $1.5 billion AI implementation company focused on helping enterprises integrate AI into core business operations rather than simply providing access to AI models. Built on the acquisition of AI engineering startup Fractional AI, Ode employs 100 engineers and works closely with Anthropic to design custom AI systems while remaining open to using competing models when needed.
The venture reflects a growing belief among leading AI companies that enterprise adoption depends on expert implementation, with OpenAI pursuing a similar strategy through The Deployment Company. → Read the full article here.
HARDWARE
OpenAI Introduces Codex Micro Keypad for Managing AI Agents
OpenAI has launched Codex Micro, a $230 limited-edition desktop keypad designed to help developers monitor and control multiple AI coding agents. Created in partnership with hardware maker Work Louder, the device features customizable backlit keys, a rotary dial, a joystick, and real-time status indicators for AI agents, including shortcuts for actions like push-to-talk and reasoning controls.
The release follows OpenAI's expansion of agent-focused tools, including ChatGPT Work, and reflects a broader shift toward dedicated hardware for managing AI workflows. → Read the full article here.
NEWS
What Else is Happening
Meta Faces AI Layoff Lawsuit: Employees claim AI unfairly flagged workers on maternity and medical leave for layoffs; Meta denies the allegations.
SpaceX Slides Back to Its IPO Price: SpaceX shares closed at $135.27, ahead of Thursday's first Starship test since May's booster failure.
ASML Raises Sales Forecast Again: ASML raised its 2026 outlook after strong AI chip demand drove better-than-expected results and customer expansion.
Microsoft Issues Record Security Fixes: Microsoft patched 570 vulnerabilities, saying AI helped uncover more flaws, including two zero-days.
OpenAI Researcher Eyes AI Drug Startup: Miles Wang is reportedly raising $200M for an AI drug discovery startup, though he disputes the details.
Tweets
Alex’s Microsoft Paint Challenge
Tell Codex to open Microsoft Paint and try to draw you pic.twitter.com/6NL1sAVvQ0
@Alex_FFView post on XPOLL RESULTS: REAL OR AI
Can You Still Tell if an Image Is AI?
Yesterday's image was REAL. Here's how you voted: It was a close call, 50.3% thought it was AI, while 49.7% correctly picked real.
 | 🟨🟨🟨🟨🟨🟨 Real (49.7%) 🟩🟩🟩🟩🟩🟩 AI (50.3%) |
