If Friday's AI Summit in San Francisco proved anything, it's this: the center of gravity in the global AI arms race just shifted east.
South Korean President Lee Jae Myung walked into the Midway event venue in Dogpatch with a national checkbook and a vision — and walked out with $950 billion in confirmed AI initiatives spanning semiconductors, data centers, physical AI, and next-generation memory. The San Francisco AI Declaration wasn't just political theater. It was a declaration of war for AI dominance.
Here's what happened and why it matters for anyone building at the intersection of infrastructure and intelligence.
The Big Numbers That Matter
Let's cut through the noise. $950 billion breaks down into two monster deals:
- SK Group & Nvidia: $500B+ — A comprehensive partnership covering AI factories, next-gen HBM4 memory from SK Hynix, and a 2-gigawatt data center from SK Telecom powered by Nvidia's Vera Rubin platform (due online 2027).
- Samsung & Broadcom: Up to $200B — A sweeping MoU covering memory chips, sub-2-nanometer foundry services, and advanced packaging for Broadcom's next-gen AI accelerator.
And this is on top of the $576 billion in state-driven semiconductor and AI projects Lee unveiled just last month. Combined, South Korea is committing well over $1.5 trillion to AI infrastructure in under 60 days.
To put that in perspective: that's more than most countries' annual GDP.
The Dinner Table That Mattered More Than the Summit
Before the main event, President Lee held private meetings with a lineup that reads like a who's-who of the AI world: Nvidia's Jensen Huang, OpenAI's Sam Altman, Anthropic's Dario Amodei, and Broadcom's Hock Tan.
The Huang meeting was especially memorable. Lee opened by referencing reports of Huang eating beer and fried chicken in Seoul — joking he'd hoped to join. Huang's response captured the mood perfectly:
"From the fried chicken restaurant to the Korean barbecue restaurant, the enthusiasm was incredible. Everyone in Korea loves AI."
"From AI chips and physical AI to AI infrastructure, Korea has achieved remarkable results under your leadership. This is truly the beginning of a golden age for Korea."
Lee's reply? "I hope this is only the beginning."
And that's the real story here. South Korea isn't just buying AI infrastructure. It's positioning itself as the factory floor of the global AI economy.
Memory Lockdown: Why HBM4 Is the New Oil
The centerpiece of the Nvidia-SK deal isn't the dollar figure — it's the memory supply chain. Nvidia effectively locked down SK Hynix's entire HBM (High Bandwidth Memory) output for the foreseeable future. HBM4, the next-generation memory standard, is the bottleneck every AI company is fighting over. Without it, even the most powerful GPUs are starved.
SK Chairman Chey Tae-won dropped the most telling line of the summit:
"I wouldn't be surprised if, the next time I see [Huang], he'll tell me those figures were too low and say, 'I want more.'"
He also revealed that Anthropic is regularly checking on future chip availability — and the company is exploring building its own chips. Meanwhile, OpenAI has directly asked SK to supply chips to the ChatGPT maker.
The message is clear: every AI lab is terrified of being left without compute.
Beyond Chips: Physical AI and Academic Pipelines
This wasn't just a semiconductor shopping spree. The summit laid out a full-stack national AI strategy:
NAVER's AI Factory — The Korean internet giant, together with Brookfield and Nvidia, is expanding its GAK Sejong data center to 200 megawatts (~100,000 GPUs on Vera Rubin). That's more than tripling the initial buildout announced in June.
Hyundai's Physical AI — Hyundai Motor Group announced a physical AI strategy anchored by a robot reference platform built with Nvidia, extending a deployment of 50,000 Blackwell GPUs. They're also integrating Nvidia DRIVE Hyperion into their vehicle platforms for autonomous driving. Robots + self-driving + AI factories. This is a full-stack play.
KAIST Joint AI Lab — Nvidia and KAIST launched the first joint AI lab between any Korean university and a global tech company, focused on agentic AI. It's backed by a multiyear commitment to fund and train Korean researchers, with access to Nvidia's full-stack AI expertise and Nemotron open models.
Seoul National University — Another NVAITC (AI Technology Center) is in the works, covering foundation models, accelerated computing, and physical AI.
What This Means for the Rest of Us
For enterprises, startups, and builders in the Middle East and beyond, this reshapes the landscape in three ways:
-
The compute crunch isn't easing. If Korea's $1.5T+ bet can't keep up with demand, nothing else will. Plan your AI infrastructure strategy assuming GPU/memory scarcity for at least the next 24 months.
-
Supply chain diversification is real. The US-China chip war drove this. South Korea is emerging as the neutral powerhouse — aligned with the West but with its own industrial base. That's a hedge worth watching.
-
Open-weight vs. closed-model debate just got louder. With Nvidia, Samsung, and SK all backing open-model research via KAIST and SNU, expect the "open-weight AI" camp to gain serious industrial muscle. The 25 US tech companies that signed this week's letter against restricting open-weight models just got powerful Asian allies.
The Bottom Line
President Lee ended his summit appearance with a simple pitch: South Korea's "dynamic AI ecosystem" is open for business. But after this week, it's clear the country isn't just a participant in the AI revolution.
It's building the engine room.
Whether you're a founder planning your next GPU cluster, a CTO evaluating supply chain risk, or just watching from the sidelines — take note. The $950 billion question isn't whether Korea can deliver. It's whether the rest of the world can keep up.
At aratech, we help businesses navigate the AI infrastructure landscape — from local AI server deployment to full-stack agentic systems. Get in touch if you're planning your next move.