🏭 AI's Supply Chain Is Breaking

Five people who touch every layer of the AI supply chain gathered at the Milken Global Conference in Beverly Hills to describe the constraints closing in on the industry. ASML CEO Christophe Fouquet said that despite record chip manufacturing investment, the market will be supply-limited for the next two to five years.

Google Cloud COO Francis deSouza confirmed the gap: Cloud revenue crossed $20 billion last quarter at 63% growth, while the committed-but-undelivered backlog nearly doubled in a single quarter.

Key Insights:

• Supply-limited for 2-5 years - ASML CEO Christophe Fouquet said chip manufacturing cannot keep pace with hyperscaler demand for the next two to five years, regardless of current investment

• Google Cloud backlog doubled - Cloud crossed $20 billion in quarterly revenue at 63% growth while committed-but-undelivered contracts nearly doubled in a single quarter

• Physical AI reaches $15B - Applied Intuition CEO Qasar Younis, whose company is valued at $15 billion, described physical AI expanding from automotive simulation into defense applications

• Search-to-agents shift - Perplexity CBO Dimitry Shevelenko described the company's move from AI-native search toward agents that take actions rather than return results

• Architecture under challenge - Quantum physicist Eve Bodnia left academia to found Logical Intelligence, challenging the foundational architecture most of the industry treats as settled

The Bigger Picture: The chip shortage isn't a supply chain inconvenience - it's a ceiling on AI growth. ASML controls the machines that make the chips; when its CEO says demand will exceed supply for five years, that's order book reality. Companies with locked-in compute agreements win; everyone else queues. The physicist questioning the entire architecture is the most important signal in the room.

Read More →

📱 Apple Breaks Its AI Lock-In

Apple is preparing to let iPhone, iPad, and Mac users choose their preferred AI model for all Apple Intelligence features in iOS 27, iPadOS 27, and macOS 27. The new capability, called Extensions, will allow providers like Google and Anthropic to power Siri, Writing Tools, and Image Playground if they add support through App Store apps.

Apple is already testing Gemini and Claude integrations internally. ChatGPT has been the only third-party AI model in Apple Intelligence since launch. With CEO Tim Cook stepping down, incoming chief John Ternus defines Apple's AI direction.

Key Insights:

• Extensions launches this fall - Users will pick any third-party AI model as their system-wide default across iOS 27, iPadOS 27, and macOS 27 through a feature Apple calls Extensions

• Google and Anthropic already tested - Apple is internally testing Gemini and Claude as replacement AI models for Siri and other Apple Intelligence writing and image features

• ChatGPT was the only option - OpenAI's ChatGPT has been the sole third-party AI model available in Apple Intelligence since launch, and that is about to change

• Different voices per model - Users will reportedly assign separate Siri voices to different AI providers so they can tell which model is responding at any moment

• Leadership transition adds stakes - CEO Tim Cook is stepping down, leaving incoming leader John Ternus to define how aggressively Apple competes in AI going forward

The Bigger Picture: This is the biggest threat to OpenAI's guaranteed slot in every iPhone. ChatGPT had the only seat at Apple's table. Opening Extensions to Anthropic, Google, and others turns Apple's 1.2 billion active devices into the world's largest AI model distribution channel - where the best-performing model wins the default. Every AI company now has a reason to compete for that spot. For Apple, aligning with the best model rather than one partner protects user experience while the new CEO figures out whether to build or partner.

Read More →

🐉 DeepSeek Rockets to $45B Round

DeepSeek is raising its first-ever venture capital round, and its valuation has surged from $20 billion to $45 billion in weeks, according to reports from the Financial Times and Bloomberg.

The Chinese AI lab, founded by hedge fund billionaire Liang Wenfeng, who controls nearly 90% of the company, had never sought outside investment before this round.

The motivation is talent retention: with rivals poaching DeepSeek's researchers, Liang opted to raise funds to offer employee equity. China's state investment fund is leading the raise, with Tencent and Alibaba reportedly also in talks. DeepSeek's V4 models remain open-weight on Hugging Face.

Key Insights:

• Valuation nearly tripled - DeepSeek's valuation jumped from $20 billion to $45 billion in just weeks, driven by investor competition to buy into the leading open-weight AI lab

• First VC round ever - Founder Liang Wenfeng controls nearly 90% of the company and had never sought outside capital before, making this a historic first raise

• State fund leads the round - China's Integrated Circuit Industry Investment Fund is leading the raise, with Tencent and Alibaba also reportedly in discussions to participate

• Talent retention drives the raise - The decision to take funding was triggered by competitors poaching DeepSeek's research staff, requiring equity to keep the team intact

• V4 stays open and competitive - DeepSeek's V4-Pro continues to rank among the top open-weight models and is freely available on Hugging Face, optimized for Huawei chips

The Bigger Picture: DeepSeek has been the AI story Silicon Valley did not want to acknowledge: a Chinese lab matching frontier performance with no VC backing. Now it is worth $45 billion before even closing a round. The state investment fund leadership makes this geopolitical as much as financial - China is funding the lab that proved American AI pricing was not justified by cost. The Huawei chip optimization is the detail that matters most: it proves DeepSeek is building a full supply chain that does not depend on US hardware.

Read More →

🧠 LeCun Bets $1B Against LLMs

Yann LeCun's AMI Labs launched with a $1.03 billion seed round at a $3.5 billion valuation, betting on world models over language models. LeCun left Meta after 12 years to build AI that understands physical reality, targeting manufacturing, robotics, and healthcare. Backers include Nvidia, Samsung, Bezos Expeditions, and Eric Schmidt.

Read More →

⚙️ OpenAI Shares AI Training Blueprint

OpenAI released MRC, a new networking protocol for large AI training clusters, through the Open Compute Project for any company to use. Developed with AMD, Broadcom, Intel, Microsoft, and NVIDIA, MRC cuts network failures and improves GPU coordination. OpenAI says more than 900 million people use ChatGPT every week.

Read More →

🔬 ChatGPT's New Default Hallucinates Less

OpenAI replaced ChatGPT's default model with GPT-5.5 Instant, which the company says produces fewer hallucinated claims than its predecessor GPT-5.3. OpenAI reports the new default delivers smarter, more concise answers with improved personalization for all users. GPT-5.5 Instant is available across free and paid ChatGPT tiers.

Read More →

Save to Spotify 🎙️

Spotify's new Save to Spotify tool connects AI agents - including OpenClaw, Claude Code, and OpenAI Codex - directly to your podcast feed. Generate an audio summary or personal podcast with your agent, add 'save to Spotify' to your prompt, and the file appears alongside your regular shows. Download the CLI from GitHub to get started.

Try it →

Google Health Coach 🩺

Google's Gemini-powered Health Coach launches May 19 at $9.99 per month, acting as a fitness coach, sleep advisor, and wellness guide in one app. Set your health goals and routine during onboarding, then log workouts and meals by voice or photo. Google AI Pro and Ultra subscribers get it included at no extra cost.

Try it →