OpenAI’s nonprofit cage is gone.

With this week’s restructure, Sam Altman now has a straight line to the public markets. The company can now raise freely, IPO when it wants, and act like what it has effectively been for years:

A capital markets machine disguised as an AI lab.

What comes next could make everything we’ve seen so far look like a warmup.

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.

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, click here to fix that.

OpenAI has finally broken free from the governance maze that’s defined it since 2019.

The company has restructured into OpenAI Group PBC. The nonprofit foundation still controls the new public benefit corporation but is now free to raise capital, pursue an IPO, and decouple from Microsoft’s exclusive compute requirements. Microsoft retains a 27% stake, worth roughly $135 billion, and a long-term cloud contract worth $250 billion, but loses its right of first refusal on OpenAI’s compute deals. Sam Altman remains CEO, without equity, and confirmed the company’s financial obligations now total $1.4 trillion across plans to build 30GW of data centre capacity globally. That includes a proposed roadmap to add 1GW per week, bringing construction costs down from $50 billion to $20 billion per site.

Why this matters:

Click here to read the full announcement from OpenAI.

Back in May 2025, the US-UAE AI pact was pitched as a cornerstone of “AI diplomacy.”

The plan? Supply hundreds of thousands of NVIDIA GPUs to power OpenAI’s Stargate campus in Abu Dhabi. 5GW of sovereign compute would be anchored by US firms like Microsoft, Oracle, and Cisco. G42, the UAE’s flagship AI group, was expected to receive roughly 20% of that chip allocation.

Then came the silence.

By mid-July, early reports suggested the deal had stalled.

Export licenses were delayed. Washington was quietly re-evaluating G42’s eligibility. The concern was simple but serious: how much US technology could ultimately end up in China’s hands. On October 9, Bloomberg confirmed the first cracks. The Commerce Department approved several billion dollars’ worth of NVIDIA chip exports to US companies operating in the UAE.

But not to local customers like G42.

Only American-run facilities made the cut.

G42’s share of the shipments was left in limbo. Two weeks later, the FT’s investigation explained why. US intelligence agencies reportedly discovered that G42 technology had been passed to Huawei, allegedly helping China upgrade the range of its PL-15 and PL-17 air-to-air missiles. The allegations, disputed by both G42 and Huawei, set off alarm bells across Washington, with some officials calling it “a red alert.”

Those revelations explain the slow pace of chip license approvals.

Despite Trump officials’ desire to push ahead, security hawks remain uneasy about exporting US AI hardware to a nation still viewed as a strategic bridge to China.

Why this matters:

Click here to read the full story from the FT.

At this year’s FII, HUMAIN and Qualcomm just signed a landmark deal.

The plan? Deploy 200MW of AI infrastructure across KSA from 2026. Use Qualcomm’s AI200 and AI250 rack systems. Deliver global inference services from Saudi soil. Anchor the Kingdom’s next growth chapter around compute, not crude.

The timing tracks with a bigger economic shift.

As Bloomberg reported last week, Riyadh is quietly scaling back megaprojects like Neom’s The Line and redirecting capital into faster-payback, technology-driven sectors.

AI. Semiconductors. Gaming. Digital manufacturing. All potentially major growth industries of the 21st Century, and doubly important with oil prices flat and fiscal pressure rising.

Saudi Arabia’s pivot from oil to intelligence is accelerating.

And with CEO Tareq Amin also pledging not to use Huawei à la G42, the race between the UAE and KSA is really heating up.

Why this matters:

Click here to read the full announcement from Qualcomm.

A new open-source contender just entered the protein-structure race.

The OpenFold Consortium has released OpenFold3, an early-access, protein-folding AI model. For readers unfamiliar, DeepMind’s AlphaFold cracked one of biology’s hardest problems in 2020: predicting a protein’s structure from its amino acid sequence. The breakthrough transformed drug discovery, vaccine design, and materials science.

But AlphaFold’s latest version, AlphaFold3, remains partly closed to commercial use, limiting who can build on it.

That’s where OpenFold3 comes in.

Developed by a non-profit collective of academic and private labs, it’s trained on 300,000 known protein structures and a synthetic database of 40 million more. OpenFold3 cost $17 million to develop, and can model how proteins interact with other molecules, such as drugs, DNA, or RNA. Future updates will expand into binding affinity and thermal stability predictions, making it a powerful foundation for biotech and pharma innovation.

Why this matters:

Click here to read the full article from Nature.

SambaNova Systems is exploring a sale after failing to close a new funding round, according to The Information.

Once valued at $5 billion, the Palo Alto firm has reportedly hired an investment bank to manage the process, but may still back away. Founded in 2017, SambaNova has raised $1.1 billion from GV, Intel Capital, and SoftBank’s Vision Fund. The company claims its SN40L AI chip delivers better performance than NVIDIA’s GPUs at a fraction of the power. Yet after struggling to gain traction, and with investors like BlackRock already marking down its valuation to around $2.4 billion, the market clearly thinks otherwise. That’s why a sale looks attractive.

The potential buyer could be a familiar name.

Bloomberg reports that Intel is in preliminary talks for the acquisition.

Intel CEO Lip-Bu Tan is well-acquainted with SambaNova. He’s the company’s executive chairman and an early backer through his firm Walden International, which led SambaNova’s $56 million Series A back in 2018. If the deal goes through, Intel could use SambaNova to shore up its AI credentials and fill gaps in its data-centre roadmap following delays to Gaudi 3 and ongoing pressure from competitors.

Why this matters:

Click here to read the full story from The Information.

Amazon is cutting 14,000 corporate jobs, its second-largest round since 2022, as it restructures around AI efficiency.

The layoffs hit roughly 4% of Amazon’s corporate staff, including roles across HR, devices, operations, and AWS. In a memo to employees, senior VP Beth Galetti said the cuts are designed to make Amazon “leaner” and “faster,” with more resources flowing toward the company’s “biggest bets”: AI and cloud infrastructure. That direction has been clear all year. Now, the wider organisation is finally feeling the inevitable effects of such massive, sustained capex without a clear ROI timeline.

Why this matters:

Click here to read the full story from TechCrunch.

China just dropped its newest AI infrastructure experiment. Straight into the ocean.

Shanghai has completed phase one of the world’s first underwater, wind-powered data centre. The $226 million (RMB 1.6 billion) project in the Lingang Special Area is designed to merge renewable energy with extreme-efficiency cooling. The 24 MW phase one installation draws over 95% of its electricity from offshore wind and uses seawater for passive cooling, cutting energy demand for temperature control to under 10%. Its target PUE is 1.15, well below China’s national efficiency target of 1.25 by year-end. Phase two aims to scale the concept to 500 MW, backed by Shenzhen Hicloud Technology, Shenergy Group, China Telecom, and INESA.

Why this matters:

Click here to read the full story from Wired.

The guardrails are off.

OpenAI’s restructure marks a turning point. Not just for one company, but for the economics of AI itself. A $1 trillion IPO is no longer mere fantasy. It’s a very real financing mechanism to bankroll the physical impact AI is having on the world.

To put it another way, AI needs power, land, chips, and steel on a scale never before seen, and fast.

At the same time, the rest of the map is shifting.

The “AI diplomacy” between the US and UAE is under review. Saudi Arabia is building its own stack from the ground up, seeking to avoid Huawei-linked missteps while weaning itself off of hydrocarbons. China is experimenting underwater. Amazon’s cost-cutting shows what happens when AI scale meets human redundancy. Open-source models continue to seek a quiet rebalancing of power, regardless of industry.

The next phase of this race won’t be about who builds the smartest models, but who can fund, power, and deploy them faster, without running out of capital.

Or trust.

See you next week.