“We’ve a great game, a noble game, before us.”

From China-compliant GPUs to open-source model wars and cross-border cloud grabs, this week was a reminder that AI is the Great Game of this generation.

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:

• NVIDIA’s $44B Quarter and $4.5B China Hit

• Saudi Launches $10B Fund to Anchor AI Empire

• New Hires, M&A, and Expansion for the Neoclouds 

• DeepSeek Chases OpenAI, Google, and Meta

• Export Limits Breed New Chips

• Alibaba Targets the Global South

• Blue Owl Capital Secures $750m for CoreWeave Build

Let’s get into it.

The GPU Audio Companion Issue #44

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.

NVIDIA’s $44B Quarter and $4.5B China Hit

NVIDIA just posted the largest quarterly result in AI infrastructure history.

Q1 revenues stand at $44.1B, up 12% from last quarter and 69% year-on-year. Net income dropped to $18.8B, down 15% from Q4, but given Blackwell demand, it’s hard not to remain bullish. GB200 NVL72 systems are rolling out across hyperscalers and neoclouds alike, and Jensen says inference workloads have grown 10x in a year. And that’s before agentic workloads even start to scale.

With results like this, it seems as though rumours of an infrastructure bubble may have been greatly exaggerated.

Why this matters:

• NVIDIA is now a $44B-a-quarter compute company, with AI infra making up 89% of sales.

• It wasn’t all smooth, with last-minute US export licensing changes triggering a $4.5B write-down on now-restricted H20 inventory.

• If you strip that out, margins would’ve hit 71.3%. Instead, they only landed at 61%. Spare a thought for the shareholders.

Read the full announcement from NVIDIA here.

Saudi Launches $10B Fund to Anchor AI Empire

Saudi Arabia launched Humain Ventures, a $10B AI fund targeting startups across the US, Europe, and Asia.

It’s the investment arm of Humain, the also-recently-announced PIF-backed AI infrastructure company tasked with turning the kingdom into a global AI heavyweight. But the capital is only one part of the play. Humain’s broader strategy includes:

• 6.6GW of planned data centre capacity by 2034 (starting with 50MW powered by 18,000 NVIDIA GPUs)

• $23B in signed infrastructure deals with NVIDIA, AMD, AWS, and Qualcomm

• A Qualcomm chip design hub in Riyadh employing 500 engineers

• Ongoing equity talks with OpenAI, xAI, and Andreessen Horowitz

• Function as a soft power VC for US and frontier AI players.

Humain is betting that whoever controls the infrastructure and flow of capital will control the future of AI.

Why this matters:

• Washington and Beijing’s new Tournament of Shadows is semiconductor geopolitics, and the US-Saudi Investment Forum reset the tone on chip exports and tech partnerships.

• The US needs capital to support initiatives like the Stargate Project, and the Gulf states need advanced semiconductors to diversify their economies away from oil dominance.

• Through aligning US corporate and geopolitical interests with Gulf capital, Washington crowds out Chinese regional influence, and Saudi Arabia inserts itself at the heart of the AI race, becoming a kingmaker in the process.

Read the full story here.

New Hires, M&A, and Expansion for the Neoclouds 

Neocloud expansion takes many forms this week.

Form one: new hires. CoreWeave just brought on Carl Holshouser (formerly Visa and TechNet) as VP of Government Affairs. Nebius, meanwhile, landed former Twilio and Cloudflare revenue chief Marc Boroditsky as CRO. He helped take Twilio from ~$400M to $4B in revenue. Now he’s tasked with doing the same for Nebius, one of only a few NVIDIA Cloud Reference Partners, and still semi-flying under the radar.

Form two?

Rumblings of M&A in Europe. Northern Data quietly let the market know that it’s received “expressions of interest from US-listed companies to enter into discussions focused on merging or acquiring its Taiga Cloud and Ardent divisions”. No decisions have yet been made, and a transaction may not come to pass. Any acquisition by a US-listed firm would also likely raise regulatory eyebrows in Brussels, especially given increasing scrutiny of AI infrastructure sovereignty.

Why this matters:

• CoreWeave and Nebius are building serious operator benches, showing that the big players are rapidly moving away from bare metal towards enterprise customers.

• Northern Data’s board and management have a duty to maximise shareholder value. 

• If a merger and resultant European market entry for the mystery US player is the best way to do that, consider Europe’s neocloud ecosystem now on notice.

Read the CoreWeave announcement here.

Read the Nebius announcement here.

Read the Northern Data announcement here.

DeepSeek Chases OpenAI, Google, and Meta

DeepSeek just updated its flagship RI reasoning model on Hugging Face.

The “minor” upgrade delivers massively enhanced depth of reasoning and inference capabilities. But, that’s not all. The accompanying lightweight release runs on a single 40–80 GB GPU with a full MIT opensource license. And with R1 overall performance now approaching O3 and Gemini 2.5 Pro levels, it’s understandable why US lawmakers keep raising alarms about open-access AI capabilities. What remains unclear is how Beijing views DeepSeek’s open model ambitions, especially as Washington seeks to hamstring Chinese ambitions.

Why it matters

• Permissive (read that as real) open-source licensing, plus single-GPU support, opens the gate to hobbyists and small teams operating with limited budget.

• Meta’s relationship with the opensource community is already somewhat sour, given the apparent benchmark gaming with the disappointing Llama 4 family, and the fact that the Llama family of models are not actually opensource.

• If DeepSeek keeps putting out more capable models, Meta no longer dominating the opensource community is an increasingly likely reality.

Read the DeepSeek announcement on HuggingFace here.

Export Limits Breed New Chips

AMD and NVIDIA are rolling out new export-compliant chips for the Chinese market, redesigned to survive US trade rules. 

AMD’s upcoming Radeon AI PRO R9700, slated for Q3 2025, is built for local AI inference and tuned for multi-GPU scalability. NVIDIA’s Blackwell-based B20 follows a similar playbook. Built on Blackwell, but with performance scaled to avoid triggering US export controls.

Why this matters:

• ~24% of AMD’s revenue, and ~12% of NVIDIA’s, come from China.

• These new China chips show that both giants see compliance as an engineering problem, not a market-ending threat, and want to maintain those revenue streams.

• Whether these export-friendly variants match the performance Chinese customers expect or simply open the door for domestic alternatives remains to be seen.

Read the full AMD story here.

Read the full NVIDIA story here.

Alibaba Targets the Global South

Alibaba is quietly redrawing the global AI cloud map.

Alibaba Cloud just expanded its global footprint to 87 availability zones across 29 regions. Now, CEO Eddie Wu is going bigger, accelerating buildout plans across Japan, Korea, Southeast Asia, the Middle East, Europe, and the Americas. 

It’s all part of a long-term bet: a “unified global cloud network” that delivers consistent AI services for Chinese enterprises, no matter where they operate. This comes as Microsoft launches its first Malaysian region (with another likely coming in Johor). But, Alibaba’s already there, and everywhere else Beijing’s Belt and Road ambitions point.

Why this matters:

• Alibaba is banking on a first-mover advantage in underserved markets, although they’ll still have to deal with the same fragmented data laws US hyperscalers contend with.

• The unified cloud play aligns with China’s broader push to anchor its AI stack across the Global South.

• Wu’s message is clear: Chinese firms expanding abroad need infrastructure, and Alibaba intends to own that runway.

Read the full story here.

Blue Owl Capital Secures $750m for CoreWeave Build

Blue Owl Capital is quietly becoming the partner of choice for the neocloud era.

The alternative investment manager closed a $750 million round this week to expand its joint venture with CTP and PowerHouse. Why? To build AI-native campuses for CoreWeave.

The funding, part of a planned $5 billion rollout, will help complete a 120MW deployment in Chesterfield, Virginia.

It follows Blue Owl’s Crusoe joint venture last week. Plus its $7B digital infra fund. And takeover of IPI Partners (parent of Stack Infrastructure and Radius DC).

Why this matters:

• Blue Owl is now embedded across the US AI infra stack, potentially running a sustained campaign to finance the AI infra backbone.

• With Crusoe, CoreWeave, and Stack now in its orbit, Blue Owl’s playbook is clear: fund the platforms shaping the next phase of compute.

• Given the anticipated scale of both CoreWeave and Crusoe’s expansion plans, this likely won’t be the last time we see Blue Owl as part of a JV.

Read the full story here.

The Rundown

I’ve used the Great Game analogy a couple of times this week, so let me explain.

The first Great Game was about land. The second, logistics. This one? Compute.

AI is an existential priority for both Washington and Beijing, with the AI rollout providing an avenue to shore up regional influence/dominance.

This week is a great example. 

On the one hand, Alibaba is scaling a 29-region AI cloud footprint targeting the Global South, and DeepSeek’s latest open-source model makes near-bleeding-edge AI accessible to capital-constrained users. On the other, Microsoft is expanding its footprint in Malaysia, and the US government ensures NVIDIA and AMD can only supply export-compliant silicon just for the Chinese market.

And in the middle of all this, we find yet more evidence that new players have joined the game.

Riyadh entered this week with a $10B fund and enough infrastructure ambition to redraw the board.

The balance of power in AI is shifting. And the stakes are higher than 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.