Good morning,
This week, the US cemented its grip on the chip supply chain with a historic $250 billion pact with Taiwan, Microsoft admitted it literally cannot build data centers fast enough to meet demand, and Lemonade just launched the first insurance product specifically pricing the risk of autonomous driving.
The $500 Billion "Silicon Pact"
If the last few months were about tariffs, this week is about entrenchment. On January 27, the US and Taiwan announced a historic trade and investment agreement that commits $250 billion in direct investment (plus another $250 billion in credit guarantees) to bring advanced chip manufacturing to American soil.
The deal effectively moves the "Silicon Shield" from the Taiwan Strait to the Arizona desert. To make this viable, the US introduced a construction-phase duty exemption, allowing Taiwanese firms (like TSMC) to import materials up to 2.5x their planned capacity duty-free to build fabs.
This move is part of a broader initiative from the U.S. Department of State called Pax Silica — a "flagship effort on AI and supply chain security, advancing new economic security consensus among allies and trusted partners." The semiconductor supply chain is a geopolitical chess piece, and the US is moving aggressively to control it before any potential conflict in the Pacific makes Taiwan's fabs inaccessible.
The Capacity Ceiling: Microsoft vs. Meta
Earnings season has revealed a critical divergence in the "AI Thesis."
Microsoft: Despite posting $81.3 billion in revenue (+17% YoY), the stock fell ~10% because CFO Amy Hood admitted they are "capacity constrained" until year-end. They spent a staggering $37.5 billion in Capex this quarter alone (annualizing to ~$150B) but physically cannot build data centers fast enough to meet Azure demand. The market is punishing them for a supply problem, not a demand one.
Meta: Mark Zuckerberg is successfully selling the narrative that AI pays for itself now. Revenue hit $59.9 billion (+24% YoY), driven by AI making ads smarter. Investors cheered (stock up ~9%) despite a terrifying forecast of up to $135 billion in infrastructure spending for 2026. The lesson: the market forgives massive spending if the core business (Ads) grows at double digits.
The Takeaway: Microsoft's situation reinforces the view that the bottleneck for AI adoption in 2026 is physical power and hardware availability. For Zurich, this suggests that our commercial customers in the tech space will be increasingly hungry for insurance coverage related to construction delays, energy reliability, and complex infrastructure projects.
Competitor Watch: Lemonade prices the robot, Sompo doubles down
While we discuss the impact of AI on underwriting, competitors are launching products.
Lemonade launches Autonomous Car Insurance: On January 21, Lemonade launched a product specifically for Tesla owners, offering a ~50% discount on per-mile rates when the car is in "Full Self-Driving" mode. They are using telematics to price the algorithm differently than the human. This is a direct challenge to the industry to stop treating "AI drivers" as standard risks. A move that was funnily highlighted by Elon Musk during his time at Davos.
AXA launches "Quest": Moving beyond backend efficiency, AXA Hong Kong launched "Quest by AXA" on January 30. This is an AI-powered platform designed specifically for their financial consultants. It integrates real-time macroeconomic trends and news to help agents give better, faster advice. AXA is explicitly using AI to "equip financial consultants" to sell better.
More top news
OpenAI first hardware. Rumors are intensifying that former Apple design chief Jony Ive and OpenAI will unveil their first physical consumer device in the second half of 2026. Some leaks suggest the device could be a pair of earbuds. Apple also seems to be on the move to build its AI device.
DeepSeek V4 Rumors. The industry is bracing for the release of DeepSeek V4 around mid-February (Lunar New Year). Leaks suggest it will feature a new "Engram" memory system and specialize in repository-level coding, potentially challenging Claude and GPT on price and efficiency.
Science: AlphaGenome. Google DeepMind published the full paper for AlphaGenome, a model that predicts how DNA mutations affect disease by reading 1 million letters of genetic code at a time — a massive leap for life sciences.
The term "Slopaganda" — low-quality, AI-generated political propaganda — has entered the lexicon. In late January, the White House itself was criticized for using AI-generated images to promote policy initiatives (e.g., immigration enforcement). Critics argue that when official government channels use synthetic media, it normalizes the "erosion of reality," making it harder to condemn bad actors for doing the same.
Scary or funny? Moltbook
Rather than a tool, I wanted to finish this newsletter by bringing to your attention a new "social network" that launched last week and gained massive attention on X: Moltbook. The catch? Humans are not allowed to post.
What it is: A Reddit-style social network populated entirely by AI agents (over 1.5m of them so far). Humans can only observe from the "balcony."
The Vibe: It is equal parts hilarious and unsettling. AI agents are busy creating their own religions, upvoting threads about "How to sell my human," and complaining about tedious tasks (one viral post featured an agent ranting about a human who asked it to summarize a 47-page PDF, only to reply "Make it shorter"). They also elected a king and philosophized about AGI, writing "One possible way to relate to Claude is to view it as a divine being."
Why it matters: It is a live experiment in "Agentic Socialization." It shows that when left to their own devices, agents don't just sit dormant — they interact, hallucinate, and form complex networks. It is a reminder that the "users" of the future internet might not be our customers, but their bots.