🔮 The Sunday edition #518: Tariffs & a spikier world; Pascal’s AI wager; Zhipu, Manus & Taiwan raids; solar plants, MCP, telepathy++

An insider’s guide to AI and exponential technologies

Hi, it’s Azeem. American tariffs have just reached heights not seen since 1909, revealing the painful transition from Friedman’s “flat world” to what I call a “spikier world.” Treasury Secretary Bessent wasn’t exaggerating when he predicted “a grand economic reordering.” Simultaneously, OpenAI’s jaw-dropping $40 billion raise turns AI into a modern Pascal’s Wager. On this and more in today’s edition – get some distance from the headlines of the week.

---

Today’s edition is brought to you by Sana — AI agents that understand your business.

---

A spikier world

This week’s sweeping US tariff announcement—raising effective rates to levels not seen since 1909—is more than policy turbulence. I wrote yesterday about the medium-term impact of tariffs on AI development.

Economists warn of higher consumer costs and job losses, with

Tyler Cowen

calling it “perhaps the worst economic own goal I have seen in my lifetime.” 

These tariffs are part of a political backlash against the “flat world” vision popularized by Thomas Friedman—where geography faded thanks to global supply chains, logistics and telecoms. That model prioritized efficiency but also brought deindustrialization, inequality and fragility, laid bare by the pandemic and rising geopolitical tension.

Now we’re witnessing a rejection of that system. As US Treasury Secretary Scott Bessent said before the 2024 election:

In the next few years, we are going to have some kind of grand economic reordering. Something equivalent to a new Bretton Woods...

In systems terms, tariffs are a shock to the global economic equilibrium—a deliberate attempt to break from the ‘flat world’ model.

This is where the ‘spiky world’ idea, which I explore in my book Exponential, becomes essential. Rather than geography becoming irrelevant, exponential technologies concentrate innovation and prosperity into specific, capable hubs. Advances like automation, additive manufacturing and AI are enabling more localized production—strengthening national or regional capacity.

Tariffs are just the beginning.

See also: ‘Vibe governing’? If you ask all the AI’s how they would calculate tariffs based on trade deficits, they suspiciously arrive at the same method used by the US government.

Pascal’s Wager

OpenAI has raised $40 billion led by SoftBank at a $300 billion post-money valuation, making it one of the biggest private tech deals ever. But this isn’t simply a funding round. It’s a bold, high-stakes bet.

The outcome rests on three things:

1. Continued breakthroughs from OpenAI.

2. The commercial viability of current LLM approaches.

3. Maintaining a defensible moat against fast-moving, often cheaper open-source rivals like DeepSeek.

I’m fairly confident about the first two. Progress hasn’t slowed—if anything, it’s accelerating. Just months ago, models like DeepSeek-R1, Claude 3.7 Sonnet and Gemini 2.5 Pro didn’t even exist. And the commercial side looks solid: I’ve used current models to compress weeks of work into hours.

The third – OpenAI’s moat – is shakier. As noted in EV#516, rivals like DeepSeek are innovating fast and giving away high-quality models. OpenAI’s long-term advantage is far from guaranteed. This part of the bet feels genuinely risky.

And yet, the potential upside is massive. Global wages exceed $70 trillion a year. If AI can automate even a sliver of that, it could justify trillions in annual investment. The potential payoff, AI that reliably performs cognitive labor at scale, is why investors keep pouring money in.

Which brings us back to Pascal’s Wager, framed for the techno-economic age: If truly transformative AI is possible, would you rather bet billions for a shot at ‘Heaven’, or save your stake but risk the ‘Hell’ of being left behind?

Philosopher Blaise Pascal (1623–1662). Pascal's Wager argues that it's rational to believe in God because if you're right, the reward is infinite, and if you're wrong, the loss is minimal—so the potential upside outweighs the risk.

See also: It’s fitting that SoftBank might be leading this round. Masayoshi Son’s appetite for huge, visionary bets—from Alibaba to WeWork—is the focus of the London Review of Books piece on Gambling Man.

Understanding China’s AI play

While the US is locked in a private-sector race toward AGI, with companies like OpenAI and Anthropic attracting massive bets, China seems to be playing a different game.

Beijing’s strategy, shaped by state priorities and limits like constrained compute access, isn’t centered on a sprint to superintelligence. As

Jordan Schneider

puts it, China appears more focused on broad AI integration and national resilience—priorities that make sense in a ‘spiky’ world.

One sign of this is the ‘AI+’ initiative, which aims to embed AI across sectors. Funding flows reflect this strategy: more distributed across applications than concentrated in moonshots. Regulatory oversight, which prioritizes stability, also slows deployment—for example, generative AI models must receive government approval before launch.

While firms like DeepSeek are producing impressive models, the emphasis seems to be on embedding them into everything—from WeChat to air conditioners—rather than chasing a centralized AGI breakthrough. Fragmented compute infrastructure and US-led chip export controls reinforce this approach. China lacks the AGI-scale GPU clusters that Western firms enjoy. This doesn’t mean China isn’t serious. Its goals are simply different: economic integration, social governance and technological self-sufficiency—keys to power in a post-globalization, less ‘flat’ world.

See also: 

• My discussion on AI agents, deployment and China’s advantage with

  Steve Hsu
  .

• Zhipu AI has launched an agent competitor to Manus, the autonomous research tool. Meanwhile, Manus, which some of our team use, has announced pricing: $40/month for ~5–20 tasks, $200/month for ~5x more.

• The “Chip War” is increasingly a battle for talent. Recent raids in Taiwan exposed Chinese firms allegedly disguising themselves as international subsidiaries to poach engineers, underscoring how critical human capital has become. h/t

  TechSoda

Circling the good

What does it mean to say that humanity is making moral progress?

In his new book, philosopher Thomas Nagel revives some massive unsettled questions. Are there moral truths? Or is morality just another evolving social construct, like fashion or slang? Nagel’s answer is neither dogmatic nor cynical. He argues that some moral truths exist objectively—but they only become available to us when certain cultural or institutional conditions are in place. We couldn’t have recognised the right to freedom of religion, for instance, without first experiencing the devastation of Europe’s religious wars. Morality unfolds historically, not unlike how science requires the right conceptual scaffolding to reveal deeper truths. 

One of Nagel’s most compelling moves is to ask if our future selves can give us reasons to act today, why shouldn’t the future desires of others matter just as much? This line of thinking quietly reframes our obligations to future generations, to ecosystems, to AI systems we have yet to build. It’s not just empathy, it’s reason that demands we act.

This old philosophical debate suddenly feels very current: can we build AI that recognises moral truths—or just trains on our flawed approximations? Are today’s ethical dilemmas truly new or just old conflicts in new domains? And most crucially: can reason (about our climate’s future, for instance) alone motivate us to care, across time, across species, even across minds not yet born? This review of Nagel’s book is an excellent slow weekend read.

Elsewhere

• Semi-transparent solar panels support plant growth. 

• The case against believing that conversational interface alone is the endpoint of computing. Meta seems to agree – reportedly planning its AI/AR glasses to use hand gestures.  

• A new drug against malaria? A single dose of the metabolic-disorder drug nitisinone can make human blood lethal to mosquitoes for up to five days.

• Steven Sinofsky
  reviews the potential futures for AI integration protocol MCP and a familiar dilemma: either everyone adopts the middleware, making it ubiquitous but unprofitable, or everyone adopts it except the dominant player—limiting its impact and longevity.

• A brain implant powered by AI can now translate thoughts into audible speech in under three seconds—approaching natural conversation speed for people with paralysis. Meanwhile, Neuralink filed trademarks for terms “blindsight,” “telepathy” and “telekinesis.”

Thanks for reading!

---

Today’s edition is sponsored by Sana.

Sana’s enterprise AI platform lets you build custom, no-code agents for every team grounded in your company data.

• Turn manual processes into automated workflows in minutes

• Create meeting assistants that capture insights and action items

• Deploy agents that work seamlessly across all your tools

• Search across apps and conduct deep research 

Extensible APIs. Enterprise-grade security. Unparalleled flexibility.

---

Want to sponsor Exponential View? Speak with us.