• The GPU
  • Posts
  • Issue #40: AI Diffusion Rules Out, Middle East In, & 5GW Campuses

Issue #40: AI Diffusion Rules Out, Middle East In, & 5GW Campuses

Feat. KSA, Humain, Nvidia, AMD, AWS, Cisco, TensorWave, OpenAI, G42, and The UAE

Author

Ben Baldieri
May 17, 2025

The AI world is no longer bipolar.

Washington. Beijing. Two heavyweights. Locked in an ideological struggle.

The prize?

Deciding how humanity’s final invention develops.

But this week saw a massive change in dynamics. A third pole has emerged: The Middle East. Abu Dhabi and Riyadh have carved out their own seats at the table and, in doing so, have rewritten the rules of engagement for everyone.

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 #40

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.

Trump Admin Tears Up AI Diffusion Rules

The US Department of Commerce has scrapped the Biden-era AI Diffusion Rule.

The Trump administration says the rule would have “downgraded allies” and burdened US companies with unnecessary compliance. Instead, the new direction emphasises surgical enforcement. Use of Huawei Ascend chips now violates US export rules worldwide, while US AI chips must not be used to train or run Chinese models. US firms are also being guided to harden their supply chains against diversion risks.

Why this matters:

  • These changes came in just days before enforcement was set to begin, giving previously second-tier allies breathing room.

  • The US stance seems to be shifting from broad containment to direct negotiations, or targeted attacks on key vulnerabilities.

  • The removal of the UAE and KSA from the Tier 2 list is central to the events of this week, like the 500k Nvidia GPUs earmarked for the UAE, Humain x AMD partnership, and everything else.

Saudi Arabia Launches Humain

Saudi Crown Prince Mohammed bin Salman has launched Humain.

Under the kingdom’s $700B+ Public Investment Fund, the new company’s goal is to turn Saudi Arabia into a global hub for AI infrastructure, data centres, and foundational models, outside of US-China dominance. Humain will offer AI services, cloud infrastructure, and advanced models. Its formation came just days ahead of the Saudi-US Investment Forum in Riyadh, where AI topped the agenda during President Trump’s Gulf visit.

Why this matters:

  • Humain signals that PIF’s next bet is sovereign AI infrastructure.

  • They’ve partnered with Nvidia, AMD, AWS, Global AI, and Cisco to deliver on their vision for KSA, which indicates a broader push to combine state capital + Western partnerships to fast-track AI industrialisation.

  • This sets up a three-way race: US hyperscalers, Chinese state-linked labs, and Middle Eastern sovereigns with cash and neutrality.

TensorWave Raises $100M

TensorWave just closed a $100M Series A to build the world’s largest liquid-cooled AMD GPU cluster of 8,192 MI325X accelerators with 256GB HBM3e each.

Unlike most NeoClouds competing in the Nvidia-saturated market, TensorWave is going all-in on AMD, both hardware and software. It’s ROCm-native, tightly optimised, and built from scratch around thermal density and open ecosystems. The bet? That serious AI builders want full control, not abstracted orchestration, and that ROCm is finally mature enough to compete head-on with CUDA. The round was co-led by Magnetar and AMD Ventures, with backing from Prosperity7, Maverick Silicon, and Nexus.

Why this matters:

  • This is the first AMD-native NeoCloud to reach true hyperscaler scale - 8,000+ GPUs with direct liquid cooling is no side project.

  • If TensorWave delivers on performance, AMD gets its best shot yet at displacing Nvidia in training and inference workloads.

  • It shows the playbook for specialist clouds. Instead of general-purpose fleets, we’re seeing the rise of tightly integrated infra tailored to specific silicon—lean, efficient, and developer-first.

OpenAI Teases New Model for Science

OpenAI chief scientist Jakub Pachocki has confirmed the company will soon release a research-focused open-weight model capable of step-by-step logical reasoning.

‘AI models are capable of novel research’: OpenAI’s chief scientist on what to expect

Jakub Pachocki, who leads the firm’s development of advanced models, is excited to release an open version to researchers.

www.nature.com/articles/d41586-025-01485-2

It’s OpenAI’s first major open-weight release in years, and it’s aimed squarely at the scientific community. These "deep research" models are designed specifically for scientific applications. Think literature synthesis, hypothesis testing, and autonomous code or hardware design generation.

Why this matters:

  • OpenAI is re-entering the open-source race on its own terms, starting with science.

  • The new model will both assist and compete with researchers, potentially accelerating scientific advancement.

  • Open-weight reasoning agents could power a new class of academic and sovereign labs without OpenAI API dependencies.

AMD Targets AI’s Quiet Gap with New CPUs

AMD launched its EPYC 4005 Series CPUs.

The first big partner? Vultr, which is deploying them to power workloads needing low latency and faster memory throughput. As NVIDIA focuses on premium GPUs, and Intel continues to stumble, AMD looks like it’s sliding into the “fit for purpose” compute layer, where most real-world inference still happens.

Why this matters:

  • Inference doesn’t always need a GPU, and AMD is quietly claiming that middle tier.

  • Vultr’s deployment shows there’s already demand.

  • AMD’s strategy is clearer than Intel’s, and potentially more relevant than Nvidia’s for AI at the edge.

AWS Commits $5.3B to Support Saudi AI Ambitions

AWS has pledged $5.3 billion to build out cloud regions and AI services in Saudi Arabia.

The investment will directly support Humain, Saudi’s sovereign AI infrastructure project. The deal gives AWS a central role in powering regional LLMs, AI data platforms, and hyperscale workloads. Under the banner of a US-aligned Gulf partner.

Why this matters:

  • With NVIDIA and AMD supplying chips into the region, and AWS and Cisco delivering the ancillary services and platforms, the US effectively embeds its AI supply chain inside a sovereign Middle Eastern deployment.

  • Humain can ride hyperscaler elasticity without ceding full control, outsourcing orchestration, not ownership.

  • By anchoring itself inside Saudi’s sovereign AI stack, AWS ensures its services become an integral substrate for Gulf model development.

UAE and US to Build 5GW AI Cluster in Abu Dhabi

The US and UAE are co-developing a 5GW AI campus in Abu Dhabi.

The project will be the world’s largest AI campus outside the US. Phase one will deliver 1GW, led by G42, and backed by a coalition of US tech partners. The campus will sit within a 10-square-mile AI zone and support sovereign compute for the Middle East, Africa, and South Asia. This is the infrastructure backbone of the US-UAE AI Acceleration Partnership, which includes chip export allowances in the hundreds of thousands, investment channels, and joint AI governance protocols.

Why this matters:

  • This is the first sovereign AI cluster with US policy pre-clearance and hyperscaler buy-in.

  • Export rules, compliance frameworks, and commercial partnerships are being designed upstream, not retrofitted after chips land.

  • With AI demand outpacing grid upgrades in the West, the UAE could also be positioning itself as a global overflow zone for high-density compute.

The Rundown

Weeks don’t get much bigger than this.

The Trump administration killing the AI Diffusion Rule and swapping it for targeted chip enforcement fundamentally changes how everyone operates in this space. Saudi Arabia’s Humain going live, with Nvidia, AMD, AWS, and Cisco lining up to form partnerships, is testament to this. The message is clear: if there’s strategic alignment with US interests, no deal is off the table. Even if that deal is as audacious as the UAE’s 5GW AI campus, and 500k GPUs.

And while these are all massive stories, that’s not all that happened this week.

TensorWave’s $19M raise to target the enterprise AI infra stack, AMD’s CPU inference offering, brought to market with Vultr, and OpenAI’s teasing of an open-weight reasoning model for scientific discovery make it seem like the game has changed. Again.

Never a dull moment.

Ever.

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.