Issue #19: Microsoft Breaks Physics, Ori Heads to KSA, and ROK Goes Beyond Hyperscale

Microsoft just made every high school physics textbook obsolete.

Satya Nadella casually announced that Microsoft had discovered a new state of matter. Because solid, liquid, gas, and plasma weren’t enough, apparently. And, naturally, they’ve used it to build Majorana 1, a quantum chip that promises to scale to a million qubits and solve problems that today’s supercomputers couldn’t even attempt.

Meanwhile, Ori just secured investment from Saudi Aramco’s VC arm, a couple of other neoclouds raised a combined $785m, South Korea is planning the world’s biggest data centre, and Intel might be up for sale.

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:

• Ori secures strategic investment to expand AI cloud in Saudi Arabia.

• Lambda lands a $480M Series D while Together AI close a $305m Series B.

• Core42 launches an AI inference playground in the UAE.

• DeepSeek’s R1 gets de-censored, and Thinking Machines goes all-in on open AI research.

• Intel faces a potential split as TSMC and Broadcom eye its assets.

• Microsoft unveils a new state of matter to accelerate quantum computing.

• South Korea announces a $35B mega data centre.

Let’s get into it.

The GPU Audio Companion Issue #19

Want the GPU breakdown without the reading? The Audio Companion does it for you—but only if you’re subscribed. Fix that here.

Saudi Aramco’s VC Arm Invests in Ori

Ori has secured a strategic investment from Wa’ed Ventures, Saudi Aramco’s VC arm.

Ori is scaling rapidly, recently becoming one of the first UK companies to deploy Nvidia’s H200 GPUs. The Saudi investment will help build a regional cloud subsidiary in Riyadh, tapping into Saudi Arabia’s Vision 2030 AI ambitions. Developing sovereign AI capability is a key consideration in the wider MENA region, and with this move, Ori seems (literally) well-positioned to capture some of this demand.

Why this matters:

• Saudi Arabia is aggressively building AI infrastructure, and Ori is now part of that play.

• MENA is becoming a major AI hub, attracting investment from global players.

• Another win for independent AI cloud providers, pushing against hyperscaler dominance.

Read the full announcement here.

Lambda Hits Series D. Together AI Land Series B

Lambda just secured $480M in Series D funding to scale its AI cloud platform.

The company is already one of the largest independent GPU cloud providers, offering high-performance compute for AI training and inference. With this funding, Lambda plans to expand its infrastructure, add more GPUs, and compete directly with the hyperscalers. Consider CoreWeave, Crusoe, AWS, and Azure on notice.

Meanwhile, Together AI closed a $305M Series B. The company is focused on providing cost-efficient AI model training and inference with a heavy emphasis on open-source alternatives to proprietary AI models.

Why this matters:

• More competition in GPUaaS likely means better developer pricing and less hyperscale dominance.

• Rounds like these show AI demand is still surging, and investor appetite is still high, even in such a crowded market.

• Both Lambda and Together AI are positioning themselves as key players in the next phase of AI adoption.

Read Lambda’s full announcement here.

Read Together AI’s full announcement here.

Core42 Expands Its AI Inference Play

Core42 has launched an Inference-as-a-Service offering powered by Qualcomm’s hardware.

The service, powered by Qualcomm’s Cloud AI 100 Ultra accelerators, is designed to cut costs, simplify infrastructure, and make AI inference more accessible for developers and SaaS providers. As Qualcomm’s AI accelerators promise less than half the cost of Nvidia-based alternatives while making large-scale AI deployments more affordable, this move could seriously shake up the inference market.

Why this matters:

• Nvidia dominates AI training, but the inference market is still anyone’s game. Qualcomm, therefore, has as good a chance as anybody of carving out a sizeable chunk.

• By adding Qualcomm to its stack, Core42 is further increasing the diversity of hardware available on its heterogeneous AI cloud, ensuring it isn’t dependent on a single vendor.

• While Qualcomm dominates mobile AI, this could be the start of a much larger push into the enterprise AI space.

Read the full story here.

DeepSeek Gets De-Censored & Mira Resurfaces

Perplexity open-sourced R1-1776, stripping DeepSeek R1’s censorship and making it one of the most open reasoning models available.

DeepSeek’s R1 was supposed to be the open-weight challenger to OpenAI. Instead, users quickly found it dodging certain topics. Especially anything related to China and the “three T’s”. Perplexity stripped those limitations, making R1-1776 a real test for what a truly open AI can be.

Meanwhile, Mira Murati (ex-OpenAI CTO) announced Thinking Machines, a lab focused on AI that puts users in control.

Murati’s new startup is taking aim at the AI customisation gap. With ex-OpenAI and Meta engineers on board, Thinking Machines is positioning itself as a serious player in user-controlled AI.

Why this matters:

• R1-1776 is a direct response to geopolitical influence in AI models.

• Big tech is doubling down on closed models, while independents are pushing open AI and user-driven control.

• Thinking Machines is another sign that OpenAI and Anthropic won’t have the US field to themselves…and likely that AI investors will throw cash at anything with the right pedigree.

Read the full Perplexity story here.

Read about Thinking Machines here.

Intel’s Future Remains in Question

Intel’s breakup might be closer than expected. And TSMC is waiting.

Despite billions in CHIPS Act subsidies, Intel’s foundry business is still struggling. The company is burning cash to compete with TSMC and Samsung, but with delays, yield issues, and an uphill battle in AI chips, investors are losing patience. Now, reports suggest a potential split, with TSMC and Broadcom eyeing up key assets. If Intel’s fabs end up under TSMC’s control, it could be a strategic hedge against Taiwan’s role as the backbone of global chip production.

Why this matters:

• Intel is barely present in the data centre GPU market, with it failing to compete with Nvidia’s dominance and AMD’s resurgence.

• TSMC absorbing Intel’s fabs could create a bulwark against Taiwan’s geopolitical risks, securing chip production outside of East Asia.

• Broadcom’s interest suggests it’s making an aggressive push into AI silicon, potentially positioning itself as a new challenger in custom AI accelerators.

Read the full story in the WSJ here.

Microsoft Makes A Quantum Leap

Microsoft just broke, or at the very least shook, physics.

Satya Nadella announced that Microsoft has discovered a new state of matter called a topoconductor, which they’ve promptly turned into a quantum chip called Majorana 1. The chip, built on a topological qubit architecture, claims to solve one of quantum computing’s biggest problems: stability. Microsoft says it now has a clear path to a million-qubit processor, years ahead of previous predictions. If true, this could make today’s most powerful supercomputers look like pocket calculators.

Why this matters:

• If Microsoft really has cracked the quantum code, this is the biggest computing breakthrough in decades, though naturally physicists are skeptical.

• If the claims are true, encryption as we know it is finished - goodbye Bitcoin and global finance.

• Microsoft has trailed behind Google and IBM in quantum computing, but this breakthrough could catapult it to the front of the race.

Read the full announcement from Microsoft here.

South Korea’s Builds Big. Very Big.

South Korea just blew the data centre arms race wide open.

A $35 billion AI data centre project, the largest in the world, is in the works at Fir Hills. This facility goes beyond hyperscale and is designed to handle next-gen AI workloads at unprecedented scale. A mix of nuclear energy, renewables, and grid-scale battery storage will power the 3GW facility. With AI demand skyrocketing, South Korea is making a play to become a major cloud and AI hub, challenging the US, China, and Middle Eastern players like the UAE and Saudi Arabia.

Why this matters:

• This project signals that the next wave of AI infrastructure won’t just be hyperscalers building in Virginia.

• As AI power demand outstrips grid capacity, more operators will look to stable, high-output energy sources like nuclear.

• This project cements Seoul’s ambitions to be a dominant force in AI, cloud computing, and digital infrastructure (while potentially playing a role in the Stargate Project).

Read the full story on DCD here.

The Rundown

Quantum breakthroughs, AI land grabs, and Intel possibly breaking apart. Just another normal week in compute.

Microsoft casually rewrote the laws of physics. Lambda and Together AI pulled in $785M combined to fuel the AI compute land grab. Core42 and Qualcomm pushed hard into inference. DeepSeek R1 got uncensored as Perplexity stripped out CCP-aligned restrictions, and the former OpenAI CTO reentered the game. Intel’s future is up in the air as TSMC and Broadcom eye its fabs. And in data centres? South Korea is going all in, planning a $35B nuclear-backed AI cluster to rival anything in the West.

Reality is optional.

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.