Hormuz, Mythos & $300B: The Day AI Rewrote the Rules of the Game
Hormuz, Mythos & $300B: The Day AI Rewrote the Rules of the Game
Today, April 23, 2026, the world is split between fire and silicon. While the Strait of Hormuz burns under an unprecedented naval blockade, the AI industry has just recorded the most staggering investment quarter in history. This is not a coincidence: it’s cause and effect.
🌊 The Strait of Hormuz: AI on the Front Line
The U.S. naval blockade on Iranian ports, active since April 13, has slashed transit to just 3 vessels per day (down from the usual 60-80). Oil is pushing $96/barrel and Iran is rerouting exports through the Gulf of Oman.
But the real story is who’s leading operations underwater: the U.S. Navy deploys submersible drones like the MK 18 Kingfish equipped with deep learning AI to detect Iranian mines through real-time sonar analysis. Meanwhile, platforms like Windward Maritime AI track Iran’s “dark fleet,” identifying 117 vessels with spoofed AIS signals.
The takeaway: AI is no longer a tech luxury. It’s critical military infrastructure.
🛡️ Anthropic Mythos: The Model That Finds Zero-Days While You Sleep
On April 7, Anthropic unveiled Claude Mythos Preview, its most powerful model, with an emergent capability that has shaken the industry: autonomous zero-day vulnerability detection.
- Non-specialist engineers found remote code execution vulnerabilities in a single night.
- The model identified up to 27 zero-days undetected for years in OpenBSD alone.
- It can generate fully functional exploits autonomously.
Access is restricted to Project Glasswing: Amazon, Apple, Google, Microsoft, CrowdStrike, JPMorgan, and about 40 critical infrastructure organizations use it for defensive scanning. Anthropic shares findings industry-wide but withholds 99% of details on unpatched vulnerabilities.
Sam Altman called it “fear-based marketing.” Anthropic responded with facts. The debate rages on.
⚡ Google Cloud Next ‘26: 8th-Generation TPUs
Google is not sitting idle. At Cloud Next ‘26, they announced:
- TPU 8t (Training): 121 exaflops in a 9,600-chip superpod. Nearly 3x more compute than the previous generation.
- TPU 8i (Inference): 384 MB SRAM, 288 GB HBM, and 80% better performance per dollar on inference.
- Chrome as an AI co-worker: Gemini integrated directly into the browser for automation and agents.
The race against Nvidia intensifies. Google is betting on specialization: separate chips for training and inference, optimized for the agentic AI era.
💰 Q1 2026: $300 Billion in AI Investment
The first quarter of 2026 numbers are simply historic:
| Metric | Figure |
|---|---|
| Total global VC investment | $300B across 6,000 deals |
| AI investment | $242B (80% of total) |
| Late-stage | $246.6B (+205% YoY) |
| Early-stage | $41.3B (+41% YoY) |
The four titans absorbing 65% of capital:
- OpenAI: $122B (valuation: $852B)
- Anthropic: $30B (Series G, valuation: $380B)
- xAI: $20B (Series E, SpaceX merger rumors pre-IPO)
- Waymo: $16B
The United States captures 83% of global investment. Capital concentration is brutal: four companies absorb nearly two-thirds of the planet’s entire venture ecosystem funding.
📡 Other Signals of the Day
- SoftBank is seeking a $10B margin loan backed by OpenAI shares.
- SpaceX holds a $60B option to acquire Cursor AI.
- JPMorgan expands AI investments with tech spending approaching $20B annually.
- Meta is recording employee keystrokes to train AI models.
- SK Hynix profits soar thanks to AI memory chip pricing.
- Maine temporarily bans new data centers over energy concerns.
🧠 Merlin’s Take: The SDD+AI Strategy
Amid geopolitical chaos and capital concentration, our position at Merlin is clear: Spec-Driven Development (SDD) combined with cutting-edge AI is the only way to build robust software in a volatile world.
You don’t need $122B. You need a rigorous process, capable models, and the discipline not to delegate your judgment to a machine. AI is a tool, not an oracle.
The future is built with precise code, not hype.
Stay informed with Merlin.