Europe’s AI infrastructure map is cracking.
The grid in traditional markets can’t keep up. Billions in data centre investment are shifting toward regions where the lights can come on fast. New analysis suggests half of future capacity could move outside the old hubs by 2035. The winners? France, Italy, and the Nordics.
The era of FLAP-D dominance is ending.
I’m Ben Baldieri. 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:
Let’s get into it.
The GPU Audio Companion Issue #48
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.
Grid Woes Reshape Europe’s Data Centre Map
No surprise here: power grids will dictate where Europe’s AI data centres actually land.
@IEA As power-hungry data centres need bigger watts and shorter waits to a grid connection, AI investments are moving🔌
By 2030, legacy hubs like Frankfurt & London will grow slower as investments shift to less congested grids, finds Ember's latest report 👇
ember-energy.org/latest-insight…
— #Ember (#@ember_energy)
8:00 AM • Jun 20, 2025
A new Ember report warns that slow grid connections in Frankfurt, London, Amsterdam, Paris, and Dublin, averaging 7-10 years, could divert billions in investment to faster-moving regions. By 2035, half of Europe’s capacity could sit outside these legacy hubs. Recent AI infra deals suggest that the shift is already well underway.
Why it matters:
Billions in GDP growth could bypass the UK, Germany, and the Netherlands, heading to France, Italy, and Scandinavia.
Policy paralysis risks turning historic data centre hubs into footnotes.
Developers will follow grid availability, not heritage.
OpenAI Lands $200M Pentagon Pilot
OpenAI is working with the US Department of Defence.
The $200m pilot is the first deal for the new “OpenAI for Government” division. The new unit aims to consolidate and expand the company’s federal work, bringing ChatGPT Gov, custom models, and hands-on support to US agencies. The flagship contract is with the DoD’s Chief Digital and AI Office (CDAO). The purpose? Using AI to streamline operations across health care access, program data analysis, acquisitions, and proactive cyber defence.
Oh, and OpenAI launched a podcast. Busy week.
Why it matters:
AI is increasingly central to modern warfighting support, moving beyond weapons to logistics, cyber, and decision systems.
The Pentagon’s $200M budget for a pilot project shows the scale of interest in frontier AI for defence efficiency and capability.
Partnerships like this could drive broader adoption of AI across the federal government, normalising its use in sensitive, high-stakes environments, echoing Palantir’s trajectory.
CoreWeave Moves Up the Stack
CoreWeave just rolled out its first joint products with Weights & Biases.
We're launching the first integrated capabilities with @weights_biases. From real-time training observability to seamless model access & live agent evaluations, AI teams can now iterate faster than ever.
coreweave.com/news/coreweave…— #CoreWeave (#@CoreWeave)
6:02 PM • Jun 18, 2025
At the Fully Connected Conference, CoreWeave made its first software move since the March W&B acquisition. The announcement? Mission Control integration for cluster observability, W&B Inference for simplified access to top open models, and W&B Weave Online Evaluations for monitoring AI agents in production. This bridges metal-level observability with app-level insights, giving customers what CoreWeave calls “metal-to-token” transparency. With Crusoe making moves in managed services as well, competition is rapidly moving away from just bare metal.
Why it matters:
CoreWeave is moving beyond raw GPUs into differentiated services because that’s where the margin (and customer stickiness) lives.
The blend of infra + tools reduces friction for AI teams, speeding up iteration and production deployments.
The neoclouds seem intent on owning more of the AI developer experience, challenging hyperscalers not just on price/performance, but on platform completeness.
Midjourney Enters AI Video Race With V1
Midjourney has launched its first AI video model.
Take a look at Midjourney's much-anticipated AI video generation model, V1, which just launched for its users at $10 a month.
V1 is an image-to-video model, so if you upload an image, it will create a series of short videos based on the initial image. And the model launch now
— #TechCrunch (#@TechCrunch)
10:38 PM • Jun 18, 2025
Users can now create 5-21-second video clips from images via Discord. The clips lean surreal, not photorealistic, and position Midjourney against Sora, Runway Gen 4, and Veo. But legal risks loom: Disney and Universal lawsuits allege Midjourney’s tools reproduce copyrighted characters.
Why it matters:
Midjourney is a serious player in the AI image market
Video generation is the logical next step.
If this move really is a first step toward larger-scale simulations, much larger markets (gaming, VR, immersive experiences) are there to be conquered.
AMD Chips to Power Next-Gen Xbox
Celebrating 20 years of partnership! AMD and @Xbox have redefined gaming together — and we’re just getting started. Our multi-year strategic partnership unleashes @AMDRyzen + @AMDRadeon power to co-engineer the future of gaming for every player, everywhere.
Read more from the
— #AMD (#@AMD)
2:05 PM • Jun 19, 2025
Microsoft and AMD have signed a multi-year deal for custom silicon covering next-gen Xbox consoles and cloud server blades for Xbox Cloud Gaming on Azure. That means AMD will power both living-room devices and data centre GPUs, eating further into markets that NVIDIA currently dominates.
Why it matters:
AMD is pushing deeper into workloads that NVIDIA has historically owned.
Cloud gaming equals more overall GPU market share.
With the new MI350-series GPUs and Helios rack-scale systems on the horizon, and now this, AMD is taking a serious swing at NVIDIA’s dominance.
OpenAI and Microsoft Rift Deepens
The world’s most famous AI partnership is really starting to fracture.
Exclusive: Tensions between OpenAI and Microsoft over the future of their famed AI partnership are reaching a boiling point
— #The Wall Street Journal (#@WSJ)
8:12 PM • Jun 16, 2025
OpenAI is considering antitrust action against Microsoft over contract terms, IP control, and compute lock-in. The dispute comes as OpenAI eyes converting to a for-profit PBC, and as EU governments step back from Microsoft reliance, with Germany and Denmark leading the charge. Bad for Satya, good for Linux.
Why it matters:
OpenAI independence would threaten Azure’s grip on top-tier models.
Antitrust action would fuel narratives about hyperscaler overreach amid EU sovereign shifts.
Even the tightest alliances can buckle under competing ambitions.
Hyperscalers Harvest Water from Waste Heat
Hyperscalers could be about to turn server heat into a fresh water source.
AirJoule has inked an agreement with an unnamed hyperscale operator to explore integrating its water-from-air tech into data hall designs. Backed by GE Vernova, the startup aims to have demo units ready by year-end. Given the growing pushback against data centre water use, the time is now.
Why it matters:
Server heat reuse is becoming standard, from district heating to water harvesting.
AI-era data centres are voracious water users, and solutions like this ease local strain.
Hyperscale developer buy-in signals that next-gen cooling is moving from theory to deployment, though many such pilots struggle to scale beyond demo.
The Rundown
Power is the bottleneck for AI infrastructure.
No power = no data centre = no GPUs = no training = no inference = no adoption. That’s why Europe’s AI buildout is shifting—billions in spend are already moving toward France, Italy, and the Nordics.
And beyond Europe?
The competitive map is changing just as fast.
AMD is coming at NVIDIA on every front. OpenAI is landing DoD contracts while fighting with Microsoft. CoreWeave is climbing the stack with infra + tools. Even server heat is being turned into water.
The race continues.
p.s. I’ve finally got a referral program up and running. Share your personal link below with just one person who you think will find it useful, and bag yourself a month of premium access as a thank-you!