• The GPU
  • Posts
  • Issue #41: OpenAI x Jony Ive, Core42 in Italy, and Crusoe's Next $15B

Issue #41: OpenAI x Jony Ive, Core42 in Italy, and Crusoe's Next $15B

Feat. OpenAI, io, Core42, iGenius, Crusoe, Vast Data, Nvidia, Microsoft, Google, MGX, Mistral, & Bpifrance

Author

Ben Baldieri
May 24, 2025

Platforms are how you win.

Microsoft cornered the home computing market with Windows and the PC. Apple dominates mobile with the iPhone. Google controls the web with Chrome.

But no dedicated AI platform exists yet - just tools buried inside iOS, Android, and Chrome.

That might have just changed.

But that’s not all.

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:

Let’s get into it.

The GPU Audio Companion Issue #41

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.

OpenAI moves to own the next platform

OpenAI just acquired Jony Ive’s stealth hardware company, io, in a $6.5B all-stock deal.

The merger brings the designer of the iPhone and his LoveFrom team in-house to design a new family of AI-native products. Why? To set the rules for the next platform that will serve as a “third core device a user would put on a desk after a MacBook Pro and iPhone.” A prototype is already in the works, aiming to move away from screens and redefine how humans interact with intelligent systems.

Why this matters:

  • Today, OpenAI depends on iOS and Android to reach users. That means App Store rules, revenue cuts, feature limitations, and zero control over interface defaults.

  • Whoever controls the user touchpoint controls usage, retention, and monetisation, and captures most of the value.

  • Great design, courtesy of Ive, was central to the iPhone’s success, so if OpenAI can achieve even a fraction of Apple’s success, dominance may yet not be a strong enough word.

Core42 and iGenius go exascale in Milan

Core42 and iGenius are partnering to deploy the largest AI cluster in Europe.

The G42 subsidiary will work with the Italian AI model pioneer for regulated industries to deploy thousands of NVIDIA Blackwell GPUs in Italy’s first purpose-built sovereign AI facility: Colosseum. The buildout will exceed 100 exaflops of performance and target finance, the public sector, manufacturing, and other regulated industries. Core42 will operate the infrastructure; iGenius will deliver models tuned for EU compliance.

Why this matters:

  • A project like this puts Italy on the AI map and cements Milan’s position at the top of the tier 2 data centre market in Europe.

  • G42’s overall strategy seems to be one of enablement, putting local partnerships front and centre.

  • Emirati capital is playing an increasing role not just in Europe but also in the global AI stack.

Crusoe expands Abilene AI campus to 1.2GW

Crusoe, Blue Owl Capital, and Primary Digital Infrastructure have started phase two of their $15 billion joint venture to build a 1.2GW AI data centre campus in Abilene, Texas.

The six new buildings will join two existing facilities energised in early 2025. Each building will house up to 50,000 NVIDIA GB200 NVL72 GPUs connected by a unified network fabric, and cooled using zero-water-evaporation liquid cooling. Construction currently employs 3,000 workers, and is expected to grow to 5,000 in this phase. The first two buildings are estimated to deliver a $1 billion economic impact over 20 years, which will increase significantly with this expansion.

Why this matters:

  • One of the world’s largest AI GPU clusters is coming online, built for hyperscale training workloads.

  • Crusoe is setting new benchmarks for speed, scale, and sustainability in greenfield AI infrastructure.

  • The $15 billion capital investment shows strong investor confidence in large-scale AI data centres.

VAST Data launches AI OS

VAST Data just unveiled a full-stack operating system for AI.

The new OS pulls together compute, storage, messaging, and real-time reasoning into a single distributed platform. It’s built on VAST’s Disaggregated Shared-Everything (DASE) architecture and introduces AgentEngine, a runtime for deploying and managing AI agents at scale. The platform includes native support for event processing, low-code agent workflows, and parallel tracing for system-wide observability.

Why this matters:

  • VAST wants to be the memory layer for AI. Not just storage, but reasoning, recall, and orchestration in one substrate.

  • AgentEngine brings agentic compute to the enterprise with prebuilt agents and low-code tooling.

  • This announcement is a move toward agent-native infrastructure, and as AI shifts from model-centric to agent-centric, VAST is positioning to own that future.

NVIDIA just went hard at Computex.

NVLink Fusion opens up NVIDIA’s high-bandwidth GPU interconnect to third-party silicon. This means hardware from the likes of MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence can now talk directly to NVIDIA GPUs. Fujitsu and Qualcomm are already designing CPUs around the fabric, which delivers 1.8TB/s per GPU, 14x PCIe 5.0, and plugs into NVIDIA’s rack-scale architecture alongside Spectrum-X networking and Mission Control orchestration.

Why this matters:

  • Custom CPUs and ASICs can now be fused into NVIDIA GPU fabrics at rack scale.

  • By opening NVLink, NVIDIA entrenches its GPU dominance without needing to own the whole stack.

  • This is a shot at UALink. Intel, AMD, and Broadcom are pushing open interconnects. NVLink Fusion is the counterpunch.

Microsoft and Google go agent-first

Microsoft Build and Google I/O landed back-to-back this week, and both pushed a clear message: agents are the future interface for AI.

At Microsoft Build, Satya Nadella introduced a suite of tools for building AI agents that can act autonomously across the web. Central to this is Recall, a new system built into Copilot+ PCs that captures everything you do and lets agents query your personal timeline. These memory-native devices represent a fundamental shift: moving from stateless chatbots to embedded, proactive agents.

At Google I/O, Sundar Pichai debuted Gemini 2.5 alongside Ironwood TPUs, Google’s new 42.5 exaflops pods built for inferential workloads. Google also launched Agent Mode in the Gemini app and Project Mariner tools for teaching agents to perform tasks by watching you once. Gemini is also getting deeper hooks into Search, Gmail, Docs, and Chrome, laying the groundwork for agents to act across the entire Google ecosystem.

Why this matters:

  • Both companies are repositioning the AI interface around ambient, continuous agents that handle tasks, memory, and context across time.

  • OS layers are being rebuilt to match, and custom silicon will play a significant role.

  • Microsoft is baking agents into the operating system with NPUs and Recall. Google’s Ironwood is optimised for long-context inference and agent workflows.

France goes all in on sovereign AI

MGX, Bpifrance, Mistral AI, and NVIDIA have launched a 1.4GW mega-project in Paris.

The joint venture, announced at the Choose France Summit, will build a sovereign exascale AI campus backed by France’s top research institutions, energy firms, and infrastructure players. It’s designed to handle the full AI lifecycle from model training to deployment, with support for sovereign cloud, low-carbon design, and next-gen hardware. The project is led by MGX (UAE), Bpifrance (France), Mistral AI, and NVIDIA, with support from EDF, Bouygues, École Polytechnique, RTE, and Sipartech. Construction starts in 2026.

Why this matters:

  • France is making a hard claim to lead Europe’s AI infrastructure.

  • 1.4GW is hyperscaler capacity and sets the benchmark for sovereign-scale AI builds.

  • MGX’s role, coupled with the Core42 deal in Italy, signals deepening cross-border alignment between Abu Dhabi and key European economies.

The Rundown

Two distinct flavours of massive move this week.

First, we have capital.

OpenAI is spending billions on dominating the next platform. Core42 and iGenius are deploying billions for Europe’s biggest sovereign AI cluster. Crusoe and partners are investing yet more billions in Abilene.

Then, we have enablement.

VAST Data is launching an AI OS to enable enterprise adoption. NVIDIA is throwing open NVLink to enable custom CPUs and ASICs within its ecosystem. Microsoft and Google are enabling the agentic web.

And then you have the UAE. With both.

MGX are enabling France’s 1.4GW sovereign AI megaproject, and G42 are enabling Italy’s arrival on the European AI stage. With billions.

I don’t know about you, but something seems to have changed about this market in the past month.

It’s accelerating.

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.

Ben Baldieri

Independent insights on GPU compute, AI infrastructure, and data centres. No fluff. Just sharp analysis on the trends shaping AI and digital infrastructure.

buymeacoffee.com/bbaldieri

Reply

or to participate.