AWS didn’t just tweak its compliance settings.

It stood up a completely separate company, governed by EU law, staffed by EU citizens, with its own security controls and advisory board. And critically, insulated from US oversight.

Why?

Because it had to.

Sovereignty is no longer a preference. It’s a prerequisite. And hyperscalers are being forced into structural compromises just to stay in the market.

I’m Ben Baldieri, and every week I break down the moves shaping GPU compute, AI infrastructure, and the data centres that power it all.

Here’s what’s inside this week:

• AWS Carves Out Independent EU Sovereign Cloud

• AMD acquires Brium, Deepens Open AI Software Stack

• Nebius Raises $1B in Convertible Notes

• xAI Targets $300m Raise, $113B Valuation

• Global Foundries Commits $16B to US

• Meta Secures 1.1GW of Nuclear

• Northern Data Goes Liquid in Pittsburgh

Let’s get into it.

The GPU Audio Companion Issue #45

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, fix that here.

AWS Carves Out Independent EU Sovereign Cloud

AWS has launched an entirely new EU-based company to run its Sovereign Cloud business.

New laws. New team. New structure. New strategy. And, most importantly, an independent advisory board “legally obligated to act in the best interest of the AWS European Sovereign Cloud”. That means total insulation from US oversight, independent operations, and the ability to keep running in the event of disruption. Music to the ears of many European customers worried about geopolitical volatility, I’m sure.

And it’s not just AWS.

Microsoft has pledged a 40% expansion of its European capacity. Google is doubling down on its Data Boundary and Sovereign Controls, even roping in Thales for local validation.

Why this matters:

• This wasn’t optional. It was existential. AWS is playing catch-up.

• Sovereignty is now the cost of doing business in Europe.

• A KPMG whitepaper predicted this exact outcome: structural separation of US cloud giants under local pressure.
  

Read the full story on The Register here.

AMD acquires Brium, Deepens Open AI Software Stack

AMD has snapped up Brium, adding to its string of quiet but strategic acquisitions.

Brium’s speciality? Optimising the entire inference stack before it hits the hardware. That means better out-of-the-box performance, less reliance on custom configs, and deeper support for AMD’s Instinct GPUs. In verticals like finance, healthcare, and manufacturing,  where domain-specific inference wins are starting to matter more than raw model size, this is a big deal.

Why this matters:

• Hardware doesn’t win markets. Software ecosystems do.

• AMD is building an open stack to challenge CUDA lock-in.

• The goal: out-of-the-box AI that just works, for sectors where time-to-value matters.

Read the full announcement from AMD here.

Nebius Raises $1B in Convertible Notes

Nebius just closed a $1 billion private placement.

The round is divided across two tranches of senior unsecured convertible notes, split evenly between 2029 and 2031 maturities. It’s one of the largest convertible raises in the AI infrastructure market to date, and gives the Amsterdam-headquartered firm fresh capital to push harder on GPU expansion, Blackwell deployment, and data centre footprint growth.

Why this matters:

• Nebius is chasing mid-single-digit billions in revenue within the next few years.

• This raise follows a $700M equity round in December.

• With liquidity like this, they’re rapidly emerging as a credible alternative to CoreWeave, Lambda, and the other large-scale US neoclouds.

Read the announcement from Nebius here.

xAI Targets $300m Raise, $113B Valuation

Elon Musk’s xAI is raising $300M through a secondary share sale.

The move values the company at $113B, and lets insiders cash out following xAI’s March takeover of X. It’s the latest in a flurry of Musk-led AI moves, with Microsoft now offering xAI models in Azure, Grok rolling out to Telegram’s 1B users, and a $5B loan package underway via Morgan Stanley.

Why this matters:

• Musk’s combined group now spans models, compute, distribution, and proprietary data collection (via Tesla).

• This raise validates xAI’s $80B standalone valuation with fresh secondary liquidity.

• Expect a larger investment round in the near future with fresh equity for new backers.

Read the full story in the FT here.

Global Foundries Commits $16B to US

GlobalFoundries is pumping $16B into its US operations.

The New York and Vermont fabs are set to expand, scaling advanced packaging, silicon photonics, and GaN-based power semiconductors. The move is backed by Apple, SpaceX, AMD, Qualcomm, and others. The overarching goal? Reduce China exposure and anchor AI chip supply closer to home.

Why this matters:

• This is one of the largest single-site investments in the US.

• GlobalFoundries is now a core pillar of U.S. fab sovereignty alongside Intel and TSMC Arizona.

• GF’s focus on photonics and 3D integration hits exactly where AI hardware is headed.

Read the Global Foundries announcement here.

Meta Secures 1.1GW of Nuclear

Meta is the latest hyperscale player to keep ageing nuclear assets online.

They’ve signed a deal with the Clinton Clean Energy Center to purchase 1.1GW of clean energy attributes. The deal kicks in from 2027, and replaces expiring state subsidies. It also keeps the Illinois plant online for another two decades. Constellation also plans to expand output with a 30MW uprate and is actively exploring adding an SMR to the site.

Why this matters:

• AI inference at scale needs round-the-clock power, and that means nuclear.

• Private capital is stepping in to replace public support.

• Nuclear assets are now critical AI infrastructure, not legacy power.

Read the announcement from Contellation Energy here.

Northern Data Goes Liquid in Pittsburgh

Northern Data has opened a new 20MW AI and HPC data centre in Pittsburgh.

The new facility is designed from the ground up for high-density, liquid-cooled GPU infrastructure. Operated by its Ardent Data Centres arm, the site uses RDHx and direct-to-chip cooling to push PUE down to 1.15. It’s being rolled out in phases, modular by design, with a focus on fast ramp-up and GPU readiness from day one.

Why this matters:

• Modular construction unlocks a much faster route to market and a faster time to value. 

• Operators can start generating revenue before full capacity is online.

• This is Northern Data’s blueprint for North American AI-first data centres, and part of a larger strategy to unlock over 850MW globally by 2027.

Read the announcement from Northern Data here.

The Rundown

More evidence that the winds have changed this week.

AWS just carved out a fully independent EU cloud unit. The US is securing more reshored capacity. And Nebius just raised $1B to scale its sovereign AI stack across Europe.

What’s going on?

Simple: we’re moving from global to local.

Hyperscalers are being forced into structural separation to meet sovereignty demands. Nations are going vertical, owning chips, data, energy, and cloud distribution. And nuclear is increasingly back on the menu, funded privately, to anchor round-the-clock inference.

Who wins?

It’s not the one with the flashiest model or the biggest marketing budget.

It’s the one who controls the infrastructure, where it runs, who governs it, and how it’s powered.

See you next week.

Keep The GPU Sharp and Independent

Good analysis isn’t free. And bad analysis? That’s usually paid for.

I want The GPU to stay sharp, independent, and free from corporate fluff. That means digging deeper, asking harder questions, and breaking down the world of GPU compute without a filter.

If you’ve found value in The GPU, consider upgrading to a paid subscription or supporting it below:

☕ Buy Me A Coffee → https://buymeacoffee.com/bbaldieri

It helps keep this newsletter unfiltered and worth reading.