On March 1, Iran launched kamikaze drone strikes against three AWS facilities in the UAE and Bahrain. It was the first time in military history that private data centers were deliberately targeted in armed conflict. AWS confirmed structural damage and disruptions to banking and enterprise services. Iran's stated rationale: the facilities support U.S. military operations and run Claude for intelligence tasks.

→ So what? AI infrastructure has crossed from digital risk into physical geopolitical risk. Hyperscalers building $100B+ in the Middle East now have to price war risk and geographic redundancy into every deployment decision. If your product relies on single-region cloud infrastructure in geopolitically sensitive areas, this is the moment to review your multi-region resilience strategy.

Samsung announced $73B in chip and research spending for 2026, up 22% from 2025 and ahead of TSMC's roughly $50B capex for the same period. HBM4 has entered commercial shipment. Samsung secured the Nvidia Groq 3 contract and signed a deal to supply HBM4 to AMD Instinct MI455X. CEO Jun Young-hyun: "Agentic AI is the primary driver."

→ So what? The HBM4 race is consolidating faster than expected. Two or three players will control the hardware layer for agentic AI and set global compute costs for the next three to five years. If you are pricing AI compute for products launching in 2027, your current economic assumptions will be outdated. Inference costs are still falling.

Claude Interviewer conducted 80,508 conversations across 159 countries and 70 languages in one week. It is the largest qualitative study of AI attitudes ever conducted. Top hope: professional excellence. Top fear: AI getting things wrong, ahead of job loss. The U.S., Europe, and Japan cluster at neutral or below; India and South America are notably more optimistic.

→ So what? Anthropic is building a proprietary dataset on real user AI attitudes that no other lab has, creating a durable research and product advantage. "Fear of AI making mistakes" outranking "fear of job loss" is a clear signal: if you are building AI products, invest more in trust signals and error handling than in raw performance.

Following ChatGPT Health in January and Copilot Health on March 12, Perplexity launched Perplexity Health in the U.S. with Apple Health integration, EHR access from 1.7 million providers, and connections to Fitbit and Oura. Three major platforms entering health AI in six weeks signals a platform war, not a product cycle.

MAI-Image-2 ranks fifth overall, with text rendering up 115 points. More significant: Microsoft AI CEO Mustafa Suleyman has formally shifted focus away from Copilot toward frontier model work. Microsoft is serious about reducing its dependence on OpenAI.

After 13% revenue growth, Tencent is cutting buybacks to fund model training and agentic AI. Combined with Samsung's $73B commitment: AI infrastructure spending is running at a scale the market has not fully priced in.

Jensen is repositioning Nvidia from chipmaker to AI platform operator through NemoClaw, an open-source platform for AI agents designed to reduce reliance on chip sales cycles and generate recurring revenue. The risk: staying ahead of Chinese competitors that are accelerating quickly.

OpenAI published details of an internal system that tracks misalignment risks and real-world agent behavior in tool-rich environments. It is the first time a major lab has publicly described a production-level agentic governance system. This will become an industry standard in the next 12 to 18 months.

OpenClaw went from 9K to 185K GitHub stars in 60 days — the fastest-growing repo in history.

Their docs? One person, plus Claude. They scaled to the top 1% of all Mintlify sites, shipping 24 documentation updates a day.

Scale your docs the same way

Most users open OpenClaw and send everything into one long conversation. OpenClaw loads the full context every time, mixing CRM questions, code requests, and random messages from last week. Output quality degrades over time.

Three setup changes that fix it:

Split into topic threads. In Telegram, create a group with just you and your bot, with separate topic channels: CRM, coding, research, updates. Each thread keeps its own focused context.

Delegate to sub-agents. Coding, API calls, file processing: route these to sub-agents running in the background. The main agent plans and reports, nothing else.

Schedule overnight jobs. Any repeating task (log review, inbox sort, CRM sync) runs on a schedule at night. Wake up to completed work, not a to-do list.

Techzip note Most frustration with OpenClaw does not come from model quality. It comes from using an agentic system like a chatbot. These three setup changes address exactly that.

World models are action-conditioned neural networks that let AI simulate the real world and plan ahead rather than just react. The piece maps the technology layer sitting beneath every major infrastructure bet happening right now: Samsung HBM4, NemoClaw, and Tencent's $5.2B are all bets on this capability.

Why it made the cut: Read this to understand why Samsung's $73B looks like logic, not a gamble.

Cooling accounts for 30 to 40% of data center energy use, making a country's climate a measurable technological asset. Nordic countries and Canada are becoming new strategic locations. The piece adds a deeper layer to today's Iran/AWS story: geography is already deciding who wins the AI infrastructure race.

Why it made the cut: Read this alongside the Iran strike story to understand why data centers are targets, not collateral damage.

Instead of committing to a specific year for AGI, the piece argues for working with broad timelines — a range of scenarios — to make better decisions under uncertainty. The case: leaders should hedge across both short and long-term possibilities rather than betting on a single forecast. Counterintuitive in a week full of confident, large-scale spending commitments.

Why it made the cut: Every story in today's issue is someone making a big bet. This piece asks whether they are betting the right way.

👉 Forward it to someone trying to keep up with AI.

👉 Read online: techzip.beehiiv.com

Techzip Newsletter
| Zipping what truly matters in the AI era.