1λ©΄ ν±
The big rush
Pentagon's eight, $30B Anthropic, β©60T Korea, SKT's 500B model
κΈ Yeongyang AI Association
μλν° Β· SF:// Β· μ 1νΈ
What I watched in the past 24 hours, in one word: rush. Pentagon, Anthropic, OpenAI, Huawei, Korea's government, SKT, and Google DeepMind all moved in the same week β best read as four pressures running concurrently, not one.
Outline
| Section | Highlights |
|---|---|
| A note from Yeongyang | Big Tech's rush as four concurrent pressures β data scarcity, compute capture, geopolitical bifurcation, regulatory pre-positioning |
| β‘ TL;DR | Five headline items |
| π Trend signal | Pressure matrix (4 rows) |
| β Today's big story | Korea's β©60-trillion strategic-tech commitment, β©9.9T for AI |
| π Models & releases | SKT A.X K1 (500B params) Β· Google DeepMind partnership |
| π’ Industry & business | Pentagon's eight-firm list Β· Anthropic $30B Β· OpenAI $25B Β· Huawei $12B |
| βοΈ Policy & regulation | EU AI Act Omnibus β high-risk obligations may slip to Dec 2027 |
| π°π· Korea focus | Team Korea AI first cut Β· Naver Agent N |
| π Regional & budget watch | AX-Sprint β©754B + Compute Infra β©2.08T Β· Yeongyang vegetable shipping support |
| β‘ Quick hits | NASA Claude Β· Manfred Β· Aletheia Β· DeepSeek V4 Β· Google Cloud Next |
| π Closing | Tracing how far the rush actually reaches |
A note from Yeongyang
What I watched in the past 24 hours, in one word: rush. The U.S. Department of Defense announced classified-network AI contracts with eight firms on May 1. Anthropic crossed $30B in annualized revenue, OpenAI passed $25B, and Huawei's AI chip line cleared $12B in the same week. Korea earmarked β©60 trillion ($44B) for strategic technologies over five years; SK Telecom unveiled its first 500-billion-parameter Korean foundation model; Google DeepMind signed with Korea's Ministry of Science.
Why all the hurry? A common theory holds that the largest labs are sprinting toward training on their own synthetic data β the way AlphaGo eventually trained against itself. But it's too early to conclude. The sharper academic warning runs the other way: model collapse. When a model trains on its own generated outputs, the distribution narrows and creativity decays. So the more honest read may be the opposite: Big Tech hasn't escaped human data β it has collided with the scarcity of high-quality human data.
Layer in GPU supply constraints, U.S.βChina decoupling, and incoming regulation (the EU AI Act, U.S. EO 14110, Korea's AI safety law debate) and the picture fills in. The rush is not a single story. It is four pressures running concurrently β data scarcity, compute capture, geopolitical bifurcation, and regulatory pre-positioning. Single-cause explanations will produce wrong prescriptions.
What this means for the value of human data, domain expertise, and lived experience is not uniform. Procedural, pattern-based knowledge (boilerplate code, document review) is likely to commoditize. Contextual, ethical, care-based, and creative knowledge may become a premium. Which kind of knowledge sits where will rewrite labor maps in places like Yeongyang as much as in Seoul or San Francisco.
From Yeongyang, this rush mostly stops at the provincial level. The β©60-trillion plan flows through metropolitan governments; the specific industry threshold as automatic exclusion rules below which most rural farms and small operators do not appear at all. Global concentration and intra-national regional concentration look similar but are not the same machine. Cross-border capital drives one; policy, infrastructure, and historical inertia drive the other. Same shape, different mechanism, different remedies.
We don't write to resist AI. We write to keep asking, from the periphery, the one question that should not stop β AI for whom?
β Chi-hoon Shin, Ph.D., Secretary-General
β‘ TL;DR
- Pentagon picks 8 AI firms for classified networks β Anthropic excluded by name
- Anthropic crosses $30B run-rate Β· OpenAI tops $25B Β· Huawei AI chips $12B (+60%)
- Korea earmarks β©60T for strategic tech, β©9.9T for AI alone
- SK Telecom unveils A.X K1, the country's first 500B-param model
- Google DeepMind signs strategic deal with Korea's Ministry of Science
π Trend signal β Why is Big Tech in such a hurry?
Today's stories converge on one point β but not on one cause:
| Pressure | Signal |
|---|---|
| Data scarcity | Model-collapse research and human-data shortage drive deeper reliance on curators, annotators, evaluators |
| Compute capture | NVIDIA chip constraints + U.S. export controls β Pentagon contracts, Huawei surge, capex into proprietary infrastructure |
| Bifurcation | U.S.βChina decoupling β eight-firm list vs. Anthropic exclusion, Huawei's $12B AI chip line filling the China gap |
| Pre-regulation | EU AI Act, U.S. EO, Korea's AI safety law β racing to set standards before windows close |
Single-cause explanations of the rush will produce wrong prescriptions. The civic-society question this raises is whether human knowledge becomes more valuable or more sidelined β and the answer depends on which kind of knowledge. Procedural, structured work likely commoditizes. Contextual, ethical, creative, and care-based knowledge may become premium. Where Yeongyang farmers, senior workers, and small-town experts land in this re-stratification is a question we'll keep asking.
β Today's big story
Korea bets β©60 trillion on strategic tech β the question of whose hands it lands in
On April 27, Korea's 18th Information & Communication Strategy Council approved a five-year, β©60-trillion ($44B) commitment to 55 strategic technology areas, with β©9.9 trillion ($7.3B) dedicated to artificial intelligence. The plan targets 35,000 GPUs and 12 core programs. The Ministry of Science announced that LG AI Research, SK Telecom, and Upstage cleared the first cut for "Independent AI Foundation Model" funding. Naver and NCSoft did not.
β± Time-to-arrival
- Seoul / SF Big Tech: 1 week
- Busan / Daegu mid-cap firms: 6 months
- Yeongyang farmers and small operators: distant
- What it means: Same shape as global concentration, but the mechanism is different. Global concentration is driven by border-crossing capital; intra-Korean concentration is driven by policy, infrastructure, and historical centralization. Therefore the remedies differ. Either way, the 30-employee threshold and MES-readiness criteria function as automatic exclusion.
Related yyaia perspective: The Singularity Doesn't Arrive for Everyone (Korean original; English summary on request).
π Source Β· Korea.kr Β· ~6 min
π Models & releases
SKT unveils A.X K1, a 500B-parameter Korean foundation model β β β β β
SKT, LG AI Research, and Upstage advanced; Naver and NCSoft were eliminated.
Why it matters: Korea has formally placed its bet. This round flows toward incumbents with compute and engineering depth. The watch point for the next round: whether domain-specialist or language-specific entrants find a path. If they don't, the same concentration pattern repeats inside Korea.
π Source Β· SKT Newsroom Β· ~4 min
Google DeepMind signs strategic partnership with Korea's Ministry of Science β β β ββ
A diplomatic signal aligned with Korea's "AI top-3 nation" rhetoric.
π Source Β· Google Korea Β· ~5 min
π’ Industry & business
Pentagon picks 8 AI firms for classified networks β Anthropic excluded β β β β β
OpenAI, Google, Microsoft, AWS, SpaceX, NVIDIA, Reflection, Oracle
The U.S. Department of Defense announced classified-network AI contracts with eight firms on May 1. Anthropic was conspicuously left out and has filed suit alleging retaliation. First formal entry of frontier AI into U.S. government classified infrastructure.
π Source Β· Washington Post Β· ~6 min
Anthropic at $30B run-rate Β· OpenAI at $25B Β· Huawei AI chips at $12B β β β β β
A new geography of AI revenue
Anthropic's annualized revenue crossed $30 billion (up roughly 3Γ from year-end 2025). OpenAI exceeded $25 billion and is reportedly considering a late-2026 IPO. Huawei's AI chip revenue is on track for $12B, up 60%, as orders for Ascend 950PR surge while NVIDIA stalls in China.
Why it matters: Revenue asymmetry translates directly into infrastructure, talent, and policy asymmetry. But the future of human domain knowledge is not uniform β procedural, structured tasks (code, contract review) commoditize fast; contextual, ethical, creative, and care-based domains may become premium. The civic-society question is which kind of knowledge moves where, and who decides.
Related yyaia perspective: AI Will Not Replace the Mechanic (Korean original; English summary on request).
π Source Β· Multi-source Β· ~5 min
βοΈ Policy & regulation
EU AI Act Omnibus β high-risk obligations may slip to Dec 2027 β β β β β
Trilogue 3 set for May 13, hiring & HR systems most affected
The European Commission, Council, and Parliament failed to reach agreement at the second political trilogue on April 28. A third trilogue is scheduled for May 13. The central question is whether high-risk AI obligations should slip from August 2026 to December 2027.
Why it matters: Korean firms that paused EU market entry over compliance costs gain a 16-month buffer. Note also that Big Tech's current scale-up partly reflects pre-regulatory positioning β racing to set the standards before windows close.
π Source Β· DLA Piper Β· ~7 min
π°π· Korea focus
"Team Korea AI" first cut β LG AI / SKT / Upstage advance β β β β β
The Ministry of Science's "Independent AI Foundation Model" first review advanced LG AI Research, SK Telecom, and Upstage. Naver and NCSoft did not.
π Source Β· Newsspace Β· ~5 min
Naver "Agent N" β Q1 shopping agent, Q2 "AI Tab" rollout β β β ββ
Naver is rolling out Agent N, with a Q1 shopping AI agent and a Q2 "AI Tab" launch.
π Source Β· CEO Score Daily Β· ~4 min
π Regional & budget watch
AX-Sprint (β©754B) and AI Compute Infrastructure (β©2.08T) β does it reach the periphery? β β β ββ
Korea's Ministry of Trade & Industry, jointly with 11 ministries, is launching AX-Sprint, β©754 billion to fund 246 AI applied-product programs. The Ministry of Science separately operates β©2.08 trillion for AI compute infrastructure (35,000 GPUs).
The honest read today: We won't qualify. But the monitoring begins β each year's actual delivery rate becomes the case for the next round.
An alternate path β Retrofitting legacy industries with AI is hard. But rural Korea has assets the city does not: idle farmland, low land prices, fewer regulatory entanglements, and a regulatory sandbox window local political will can open. AI-native rural industry β farming, eldercare, tourism transformed by physical AI β may begin from these conditions, with the senior farmer's domain knowledge as the most valuable input rather than the obstacle.
Related yyaia perspective: Why Yeongyang Has the Edge (Korean original; English summary on request).
π AX-Sprint Β· MOTIE Β· GPU Infra Β· NIPA Β· ~6 min
Yeongyang 2026 vegetable shipping facility support β β β ββ
On its face legacy infrastructure; through an AI-native lens, an opening for forecasting, pricing, and planning where the senior farmer is the domain expert and AI is the tool.
π Source Β· Yeongyang County Β· ~3 min
β‘ Quick hits
- NASA Perseverance β first Mars drives planned by AI (Anthropic Claude) β Nature
- Manfred β AI agent autonomously forms US LLC β ClawBank
- Aletheia agent autonomously resolves 4 open ErdΕs conjectures β arXiv
- DeepSeek V4 paper β bigger models, less compute β SCMP
- Google Cloud Next 2026 β A2A protocol, Project Mariner debut β TheNextWeb
π Closing
Tomorrow we'll trace how far this four-pressure rush moved from the provincial level downward, and where it stopped. The accumulating record is what the next round of policy debate will need to read.
β Chi-hoon Shin, Ph.D., Secretary-General
Related yyaia perspective: Investing to the Threshold β on why evidence accumulates before policy moves (Korean original; English summary on request).
yyaia curates AI signals each morning from Yeongyang, Korea. Deeper analysis or commissioned pieces β contact@yyaia.org.
Prepared with assistance from Claude / Anthropic. This issue went through a multi-layer review (Thesis β Antithesis β Synthesis) β a critic subagent surfaced six counter-arguments which the Secretary-General then integrated into the final draft. Fact collection, classification, and drafting involved AI; perspective, editorial decisions, signature, and final approval were performed directly by Secretary-General Chi-hoon Shin, Ph.D. Tool stack: yyaia.org/ai-stack.
Released under CC BY-NC 4.0.
곡μ νκΈ°