Top Story: A comprehensive long-term strategy for rare earths
❌ Claim: China got lucky sitting on all those rare earths.
✅ Fact: China’s dominance is the result of a deliberate long-term industrial policy. And what was built can be countered.
China’s decision to restrict rare earth exports last week reignited debate about resource dependency. But most people overlook three crucial points:
Rare earths aren’t actually rare. The term “rare” is a misnomer; elements like cerium are nearly as common as copper. The challenge is in extracting and purifying them, which is expensive and environmentally risky.
Processing, not reserves, underpins China’s dominance. China mines about 70% of global rare earths and controls roughly 90% of the downstream processing capacity, an edge built on lax permitting, subsidized pollution controls, and huge government investment.
Western nations can rebalance supply chains, but only with a coordinated, long-term supply and demand strategy.
What that could look like in broad strokes:
Unlock domestic sources: Speed up permitting, streamline environmental reviews, and share revenues with First Nations to strengthen trust and project viability. New projects in the US and EU take years to approve and many get stuck in US courts because they touch first nation territory but there is no framework to share benefits equitably.
Build new resource alliances: While the US has good reserves of light rare earths, we will need other partners. Vietnam and Brazil hold vast reserves — especially heavy rare earths. Strategic alliances here can reshape the global map.
Guarantee the floor for processing investments: Starting a processing plant in the US or EU is risky due to volatile offtake prices. Governments can provide minimum price or offtake guarantees to ensure a predictable ROI for processing projects.
Treat rare earths as strategic inventory: Require large manufacturers to hold minimum stockpiles — an insurance policy against geopolitical supply shocks.
The takeaway? China built its dominance by seeing decades ahead. Now, the U.S. and EU must play the long game — starting today.
Market Pulse: Key News You Need to Know
OpenAI launches “ChatGPT Atlas” — an AI-first web browser with Agent Mode
What Happened: OpenAI unveiled Atlas, a desktop browser with ChatGPT built in, “Agent Mode” for autonomous tasks, and optional long-term “browser memories.” Early coverage flags performance trade-offs. (OpenAI)
Why It Matters: The interface war is shifting into the browser itself—where control of search, shopping, and task automation lives. Security remains the elephant in the room: even OpenAI’s CISO acknowledged prompt-injection is still an unsolved problem for agentic browsing. (The Guardian)
Who It Affects: Consumers, publishers, ad platforms, and anyone building agentic workflows on the open web.
What’s Next: Watch how fast Atlas hardens its security model and improves latency; that will determine real adoption.
Read more: https://openai.com/index/introducing-chatgpt-atlas/ • https://www.reuters.com/technology/openai-unveils-ai-browser-atlas-2025-10-21/ • https://www.theguardian.com/technology/2025/oct/21/openai-chatgpt-web-browser-atlas
Anthropic doubles down on enterprise agents: Claude Code (web) + Claude Skills
What Happened: Anthropic shipped Claude Code on the web (managed, sandboxed coding agents that can work across repos) and “Claude Skills,” a way to bundle an org’s SOPs, scripts, and tools so agents execute company-specific workflows. (Anthropic)
Why It Matters: Instead of a consumer browser play, Anthropic is embedding agents inside secure enterprise workflows—prioritizing reliability, governance, and repeatability.
Who It Affects: Engineering leaders, CIOs, compliance/risk teams looking for lower-friction agent deployment.
What’s Next: Expect deeper GitHub/CI hooks and admin controls (audit, policy, approvals) as customers scale usage.
Read more: https://www.anthropic.com/news/claude-code-on-the-web • https://www.anthropic.com/news/skills
Compute arms race escalates: OpenAI–Broadcom custom silicon; Anthropic–Google Cloud 1M TPUs
What Happened: OpenAI announced a Broadcom partnership to deploy 10 GW of custom AI accelerators; Broadcom said shares jumped on the news. Anthropic inked a deal with Google for access to up to 1,000,000 TPUs and >1 GW of capacity starting in 2026. (OpenAI)
Why It Matters: Control of bespoke compute (and the power to run it) is becoming the gating factor for model roadmaps and product reliability.
Who It Affects: Model labs, cloud providers, chipmakers, and any enterprise betting on frontier-scale AI.
What’s Next: Watch delivery timelines (2026+) and interconnect/networking choices; delays ripple into model releases.
Read more: https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/ • https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-announce-strategic-collaboration-deploy-10 • https://www.anthropic.com/news/expanding-our-use-of-google-cloud-tpus-and-services • https://www.reuters.com/technology/anthropic-expand-use-google-clouds-tpu-chips-2025-10-23/
Meta trims ~600 roles in Superintelligence Labs amid push for speed
What Happened: Meta is cutting around 600 roles across FAIR, product AI, and infra groups; reporting frames it as streamlining to reduce bureaucracy and move faster. (Reuters)
Why It Matters: Highlights the trade-off many labs face: smaller, “load-bearing” teams vs. institutional risk functions—especially notable given regulatory and privacy obligations.
Who It Affects: AI researchers, safety/compliance teams, and partners reliant on Meta’s internal AI cadence.
What’s Next: Track internal reassignments and whether safety/review capacities are rebuilt in new forms.
Read more: https://www.reuters.com/business/meta-is-cutting-around-600-roles-ai-unit-axios-reports-2025-10-22/
Google turns Maps into an AI moat: Gemini API gains live Maps grounding (priced at $25/1K prompts)
What Happened: Developers can now ground Gemini responses in real-time Maps data spanning 250M+ places; pricing is $25 per 1,000 grounded prompts. (blog.google)
Why It Matters: Proprietary, fresh geospatial data is hard to replicate—this is classic data-network-effect moat building.
Who It Affects: Logistics, local search, retail, travel, and any app needing location intelligence.
What’s Next: Expect usage gating and selective enablement to control cost/performance at scale.
Read more: https://blog.google/technology/developers/grounding-google-maps-gemini-api/ • https://ai.google.dev/gemini-api/docs/maps-grounding
Global pricing geopolitics: DeepSeek undercuts on tokens; OpenAI expands “Go” across emerging markets
What Happened: DeepSeek is pushing into Africa with ultra-low model pricing (e.g., ~$1.10 per 1M output tokens), while OpenAI broadened its sub-$5 ChatGPT Go plan into new regions. (Bloomberg)
Why It Matters: Price-led expansion can seed long-term influence and data flywheels in high-growth markets.
Who It Affects: Local startups, telcos, public sector digitization, and incumbents defending ARPU.
What’s Next: Watch local currency pricing and reliability SLAs—cheap only wins if it also works.
Read more: https://www.bloomberg.com/news/features/2025-10-22/china-s-deepseek-pushes-into-africa-making-ai-accessible-to-millions • https://techcrunch.com/2025/10/09/openais-affordable-chatgpt-go-plan-expands-to-16-new-countries-in-asia/
Resilience > efficiency: AWS US-EAST-1 outage spotlights single-region risk
What Happened: A major AWS incident cascaded across thousands of services; analysis ties impact to US-EAST-1 centrality and DNS/DynamoDB issues. (theverge.com)
Why It Matters: The week’s thesis from your transcript—budget for redundancy, even if it’s “inefficient”—played out in real time. Multi-region, multi-cloud, and supplier diversification are now board-level items.
Who It Affects: Every internet-facing business.
What’s Next: Expect audits of region placement, failover drills, and new SRE budgets.
Read more: https://www.theverge.com/news/802486/aws-outage-alexa-fortnite-snapchat-offline • https://www.theguardian.com/technology/2025/oct/24/amazon-reveals-cause-of-aws-outage
