The gloves are off.

The Trump administration just revealed its plan for global AI dominance. And not just of models. Power. Chip production. Industrial applications.

Everything is on the table.

Victory is the only acceptable outcome.

This week also saw OpenAI lock arms with the UK, CoreWeave pile on more debt, Crusoe take delivery of gas turbines, and Furiosa notch a real-world GPU alternative win. Microsoft, meanwhile, is discovering that sovereignty isn’t something you can bluff.

Especially when key engineers are in China, your customers sit in Europe, and your head of legal can’t promise sovereign data remains sovereign.

I’m Ben Baldieri. Every week, I break down what’s moving in GPU compute, AI infrastructure, and the data centres that power it all.

Here’s what’s inside this week:

Let’s get into it.

The GPU Audio Companion Issue #53

Want the GPU breakdown without the reading? The Audio Companion does it for you—but only if you’re subscribed. If you can’t see it below, click here to fix that.

Trump Admin Releases AI Action Plan

The White House has just unveiled a sweeping AI Action Plan, positioning the United States for “global dominance” in artificial intelligence.

The plan, framed as a moonshot roadmap, focuses on three pillars:

Pillar 1: Accelerate AI Innovation

This pillar pushes for full-spectrum AI development and adoption. It’s pro-open source, pro-frontier model investment, and pro-productivity. Think fewer regulations, more incentives.

  • Open source support: Encourages the proliferation of open-source and open-weight models.

  • AI for industry: Incentivises AI adoption in defence, government, science, and domestic manufacturing.

  • Legal clarity: Targets legal adaptation for synthetic media and AI-driven IP to avoid chilling effects.

Pillar 2: Build American AI Infrastructure

This is the meat of the plan.

  • Energy grid expansion: Calls out the need to vastly increase U.S. energy generation, highlighting AI as a grid-scale problem.

  • Data centre growth: Targets the construction of secure, military-grade AI facilities.

  • Workforce development: Emphasises vocational and technical training to supply electricians, engineers, and operators.

  • Semiconductor reshoring: Promotes fast-track permitting and capital for fabs.

This is the closest thing we’ve seen to a national reindustrialisation strategy, built around AI.

Pillar 3: Lead in International Diplomacy & Security

This part is more adversarial.

  • Export AI, not risk: Pushes for American models, chips, and standards to be embedded in allied nations’ infrastructure.

  • Clampdown on China: Proposes tighter enforcement of export controls, as well as plugging existing loopholes.

  • Global alignment: Calls for closer coordination with allies on AI safety, biosecurity, and model evaluation.

Why this matters:

  • Sovereign AI is now the official US doctrine.

  • This is the most aggressive national AI infrastructure plan to date, and it’s part industrial strategy, part diplomatic manifesto, and part cold war playbook updated for the 21st Century.

  • This isn’t just about models. It’s about who builds the compute, owns the fabs, runs the grids, and writes the rules. Expect a fast-tracked wave of permitting, power procurement, and site approvals. Whatever it takes to win.

OpenAI Strikes Deal with UK Government

OpenAI has signed a new agreement with the UK government to explore how artificial intelligence can support national productivity.

The deal, framed as a strategic partnership rather than a binding contract, will see OpenAI work with departments across education, defence, security, and justice. It also includes plans for joint safeguards, data-sharing, and expanded investment in AI infrastructure, meaning new data centre buildouts are likely. OpenAI will also grow its London office beyond its current 100+ headcount.

Why this matters:

  • This aligns OpenAI with the UK’s AI Opportunities Action Plan, cementing the UK’s position as a preferred base for US AI firms expanding abroad.

  • Expect more infrastructure deals, as “exploring investment” usually means expanding OpenAI’s regional training and inference capacity.

  • This deal is part of a broader UK charm offensive as the government seeks to drive investment and follows on from prior deals/MOUs signed with Anthropic and Google earlier this year.

Click here to read the full story via BBC News.

CoreWeave Seeks $1.75B in Bond Sale

CoreWeave is back again.

Less than two months after announcing a $2 billion debt raise, the neocloud is seeking an additional $1.75 billion in senior notes. Why? To refinance existing debt. With $8 billion in total debt reported as of December 2024, and starting earlier this week with a $1.5 billion ask, this upsized 2031-dated bond comes as CoreWeave continues to navigate a stretched balance sheet amid rapid expansion and acquisitions.

But that’s not all that happened this week for the neoclouds.

Crusoe is doubling down on energy efficiency for AI infrastructure with a major new deal in partnership with GE Vernova. Together, they’ve announced a massive order for 29 gas turbines, aimed at powering Crusoe's growing data centre network. The focus?Combining high-performance AI infrastructure with sustainable energy sources.

Why this matters:

  • Deploying AI infrastructure is capital-intensive and power-hungry.

  • CoreWeave is arguably deploying more capacity than anyone else, and given market immaturity and a limited number of financing partners capable of providing the required liquidity, they’re saddled with high capital costs which they can now refinance through public markets.

  • Per this weeks AI Action Plan from the US Government, power is the bottleneck. This partnership with GE puts Crusoe in the enviable position of being far closer to solving it than some of their competitors, given how insane the gas turbine market is right now.

Click here to read the full CoreWeave story on DCD.

Click here to read the announcement from Crusoe.

Latent Labs Brings Protein Design to the Browser

Latent Labs has launched a web-based AI model that lets researchers design custom proteins using natural language.

The LatentX release comes six months after raising $50M and brings something genuinely new to the market. Unlike AlphaFold, which predicts existing structures, LatentX generates new ones: nanobodies, antibodies, and other synthetic molecules never seen in nature. With backers including Radical Ventures, Sofinnova, Jeff Dean (Chief Scientist, Google Deepmind/Research), and Dario Amodei (CEO & co-founder, Anthropic), and early results showing lab viability, Latent Labs could be one to watch.

Why this matters:

  • Biotech is one of the new frontiers in AI research.

  • LatentX means biotech teams can now prototype molecules without building internal models.

  • In turn, this opens the door to “foundational models” for biology, not just language.

Click here to read the full story on TechCrunch.

Furiosa AI Lands an LG-Shaped Whale

FuriosaAI has secured one of its biggest wins to date.

LG AI Research will deploy the startup’s RNGD inference chips to power its EXAONE LLMs across sectors like electronics, telecoms, and biotech. After extensive testing, LG found RNGD achieved 2.25x better performance per watt compared to GPUs while simultaneously meeting latency and throughput demands at enterprise scale.

The RNGD chips are built on Furiosa’s custom Tensor Contraction Processor architecture, with 512 TFLOPs of FP8 compute in a 180W envelope. The servers will be deployed on-prem in Korea and integrated with Furiosa’s software stack, including vLLM compatibility, Kubernetes orchestration, and OpenAI-style API support.

Why this matters:

  • This is one of the first serious GPU alternatives adopted by a major enterprise for LLM inference.

  • Power efficiency and capex control are driving buyers to consider non-NVIDIA chips, especially for sovereign AI.

  • Furiosa joins the shortlist of credible non-GPU silicon vendors landing real LLM workloads.

Click here to read the full announcement on Furiosa’s website.

Microsoft, China, the DoD, & Risky Data Access

After ProPublica revealed that Microsoft had been using engineers in China to help maintain Pentagon cloud systems, the company is finally pulling the plug.

At a high level, Microsoft has been using engineers in China to help maintain the DoD’s computer systems. The arrangement, which helped Microsoft win the federal government’s cloud computing business a decade ago, relies on US citizens with security clearances to oversee the work and serve as a barrier against espionage and sabotage. But these “digital escorts” often lack the technical expertise to police foreign engineers with far more advanced skills, ProPublica found.

This comes just days after Microsoft executives told French lawmakers they couldn’t guarantee European customer data would be safe from US legal demands.

Not a good week when you’re trying to reassure customers that they don’t need to worry about the risk of all the things that just happened happening.

Why this matters:

  • Data sovereignty is one of the most important considerations in the era of AI, and multiple European nations are seeking alternatives to US hyperscalers.

  • The France hearings showed just how fragile trust has become in hyperscaler models. Europe wants digital sovereignty. Microsoft just admitted it can’t provide it.

  • If Microsoft is seen as unable to keep the data from some of its most lucrative contracts secure, and “trust us, bro” seems to be the guardrail of choice against both geopolitical adversaries and the federal government, what hope do less valuable customers have?

Click here to read the full story on ProPublica.

Teesworks Named UK’s Second AI Growth Zone

The UK government has picked Teesworks as its second official AI Growth Zone.

The site, in the northeast of England, is the second AIGZ after Culham, and is backed by the Department for Science, Innovation and Technology (DSIT). Government officials have reportedly engaged with major firms including Anthropic, DeepMind, Microsoft, and Google. However, the decision clashes with BP’s blue hydrogen project in the same area. Developers behind the data centre argue that hydrogen infrastructure would make the build unviable, potentially impacting the UK’s energy and net-zero ambitions.

Why this matters:

  • On the one hand, AI is a hot-button political issue in the UK, and the AI Growth Zone Initiative is one of the government’s flagship industrial policies.

  • On the other hand, energy independence and the push to net-zero emissions are equally important manifesto promises upon which Labour came to power.

  • A multibillion-pound AI zone overlapping with hydrogen ambitions, water access battles, and EXTENSIVE legacy governance issues seems sadly on-brand for the UK in 2025. Ed Miliband, Secretary of State for Energy and Climate Change, has until August 28 to rule on it.

Click here to read the full story in the FT.

The Rundown

Sovereignty is the thread that ties this week together, but it looks different depending on who’s holding the needle.

For OpenAI and Oracle, it’s sovereign capacity.

5GW of American infrastructure. American power. And American jobs.

This is AI industrial policy, privately executed.

For the UK, it’s sovereign resilience.

Teesworks marks the next big move in reindustrialising regions outside the South East. Building local AI infrastructure that doesn’t live and die by London or imported clouds could reinvigorate parts of the country that have been left behind since the era of Thatcher and Reagan.

Assuming the government can make a decision.

For LG and Furiosa, it’s sovereign strategy.

A Korean global champion picking a Korean chip startup. A clear signal that walking around from Meta’s $800m buyout offer was warranted.

No GPUs, no ecosystem lock-in, no waiting on NVIDIA allocations, and further hints of alternative semiconductor viability.

And for Microsoft?

It’s sovereign theatre. Telling France it can’t protect your data. Telling the DoD it used China-based engineers, but not to worry, because the less qualified, underpaid supervisors totally knew what was going on. And telling the rest of us to pay no attention to the man behind the curtain.

Whatever keeps the contracts coming.

Sovereignty is evolving beyond strategy and into a fully-fledged business model. And it’s starting to look like some players are better at it than others.

See you next week!

Reply

or to participate