ExoBrain weekly AI news

11th April 2025: Trump hands China AI advantage, datacentres hunger for power, and Google connects agents

Author

Joel Miller
April 11, 2025

Welcome to our weekly email newsletter, a combination of thematic insights from the founders at ExoBrain, and a broader news roundup from our AI platform Exo…

Themes this week:

  • US trade policy handing China an advantage in AI development

  • AI's dual role as energy consumer and efficiency optimiser

  • Google's new A2A protocol enabling agent-to-agent communication

Trump hands China the advantage

Bill Clinton’s chief strategist James Carville famously said: “I used to think that if there was reincarnation, I wanted to come back as the President or the Pope… But now I would want to come back as the bond market. You can intimidate everybody.” His words ring truer than ever as the bond investors took aim at the foundations of US economic power. The sustained sell-off in US treasuries stems from anxiety over chaotic trade policy, vast national debt (debt-to-GDP is at 122%) and a sense that the US can’t be fully trusted. Faith in the dollar and US debt as the ultimate safe haven is decidedly “yippy”; look no further than the divergence between yields and the dollar. Typically, they move together – higher yields attract capital, boosting the currency. Yet since the latest tariff push, yields have soared while the dollar has plunged. We may be about to see a reconfiguration of global financial power if confidence continues to erode, and the outlook for US AI dominance is equally fragile.

Last week, we explored how despite exemptions, tariffs will inflate datacentre costs and slow AI compute roll-out. Despite the perceived U-turn, high tariffs targeting China persist and as of Friday the total tariff burden had actually increased (125% in both directions) on the basis of US import volumes. Now, with the trade conflict focusing squarely on China, the question is, can America maintain its AI power when its economic stability is under strain, and can it maintain a coherent and effective policy to achieve its dominance goals?

The signs are not good. Trump has a long-held belief in tariffs but perhaps an even greater love for “the deal”. The administration has shelved planned restrictions on exporting powerful new H20 Nvidia GPUs (H100 equivalents) to China after high-level lobbying, including a $1 million Mar-a-Lago dinner between Nvidia’s CEO and Trump, sweetened further by promises of US investment. Meanwhile DOGE has weakened staffing capacity to monitor critical exports, and the Biden CHIPS Act that offered a carrot, not just a stick, to encourage semiconductor on-shoring is being gutted. Regulation is under review, and the administration is pushing for more AI adoption across the federal government, but at the same time universities and research funding is under fire, and China continues to lead on churning out STEM PhDs 2:1. Perhaps in an indication of the appreciation of this area, Trump’s education secretary referred to AI as “A-one” during a recent panel discussion.

Unless a deal is done with China in the coming days (and that may be the idea) the big tech firms will continue to suffer. Furthermore, the bond market situation means additional drag on the industrial complex. Borrowing costs will rise making investments in infrastructure and R&D more expensive. With business and consumer confidence collapsing, reduced demand may also develop. DeepSeek is rumoured to be ready to drop it new model R2 in the coming weeks, if it’s as big a perceived challenge to US labs as R1, it could significantly worsen the situation for the US tech industry.

Takeaways: This is code red for US AI. Continued extreme tariffs and a recession will cripple America's ability to fund essential AI infrastructure and innovation. Simultaneously, despite tough trade rhetoric, the administration is allowing chips to flow into China, the target of their trade war. Although China will also suffer economically, it maintains greater centralised economic control and capacity to absorb setbacks. More importantly, China is ahead on self-sufficiency (AI chip self-reliance with the Huawei Ascend series and home-grown AI capabilities). It’s hard to see this contradictory, self-sabotaging strategy doing anything other than making America the main loser in its own performative game.

AI’s growing appetite and the race for clean power

The International Energy Agency’s special report on Energy and AI has laid out the most detailed picture yet of how AI is reshaping global energy systems. The central message: AI is both a new demand centre and a new optimisation tool, and we need to prepare for both roles at once.

According to the IEA, electricity consumption by data centres is set to more than double by 2030 — and AI-focused facilities will quadruple their usage. In the US, energy demands from these clusters could soon exceed those of legacy heavy industries like steel and cement. Globally, data centre electricity consumption is expected to rise from 415 terawatt-hours in 2024 to nearly 1,200 terawatt-hours by 2035.

Despite those dramatic growth curves, the IEA is careful to put things in perspective. Even at 2030 levels, data centres are likely to account for around 3% of global electricity demand. That’s still modest compared to sectors like transport or manufacturing. And much of the concern comes from where these data centres are built and how quickly grids can adapt to accommodate them — not just how much they use.

But the more interesting part of the IEA’s analysis is about AI’s positive energy impact. The report highlights how AI tools can increase efficiency across electricity networks, industrial processes, and even the built environment. The agency estimates that full deployment of today’s AI optimisation tools could cut energy-related emissions by up to 5% by 2035 — more than offsetting the emissions caused by the data centres themselves in most scenarios.

Still, long-term solutions will require breakthroughs in clean power generation. And that’s where nuclear fusion enters the conversation.

China has an increasing lead in the race to commercial fusion. Its “artificial sun” — the EAST reactor — has set endurance records for plasma generation, moving the country closer to practical fusion-based electricity. Meanwhile, US startups like Commonwealth Fusion Systems (with its MIT-linked SPARC reactor) and Helion are promising to deliver net energy fusion in the next few years but face steep challenges in scaling up and securing funding.

If fusion does arrive, the implications for AI — and energy as a whole — are huge. A virtually limitless, emissions-free energy source would decouple AI progress from today’s trade-offs in grid pressure, fossil generation, and water usage. And more broadly, it could reduce the carbon intensity of other industrial processes — including those involved in chip fabrication, steel production, and even synthetic fuel development.

Takeaways: The numbers tell an interesting story about AI's energy footprint. While data centres supporting AI systems are creating new demand at an unprecedented rate, this consumption remains a small fraction of global energy use compared to transport or traditional industry. The IEA's analysis suggests AI could actually be climate-positive if its efficiency tools see widespread adoption across energy systems, industrial processes, and transport networks. However, the long-term picture gets complicated. For AI growth to stay compatible with climate goals beyond 2030, we'll need substantial clean energy breakthroughs. Nuclear fusion, once dismissed as perpetually 30 years away, is now attracting serious investment and showing early technical promise. The race is on between AI's growing appetite for power and our ability to generate that power cleanly.

Google helps agents communicate

This image illustrates Google's new Agent-to-Agent (A2A) protocol released this week and supported by a wide group of industry partners. It shows how client agents (blue) and remote agents (green) communicate through a structured framework. The protocol enables four key capabilities: secure collaboration, task and state management for complex operations, user experience negotiation to ensure proper content delivery, and capability discovery allowing agents to find out what other agents can do. A2A represents an exciting development in interconnected multi-agent systems that can work together across the many emerging platforms.

Weekly news roundup

This week's developments show a strong focus on AI hardware investments and infrastructure scaling, while concerns about AI governance and transparency continue to shape industry discussions and regulatory approaches.

AI business news

AI governance news

AI research news

AI hardware news