SF://

slow place. fast signal.

Issued · SF:// · 2026-05-09 (KST)

---

🧭 Editor's View — How to read today's headlines

This week's headlines collapse into two sentences.

One, the assumption that AI infrastructure is locked behind a single company is starting to wobble. Apple-Intel chip talks (TechCrunch 5/8, CNBC 5/8), IREN-Nvidia 5GW DSX (CNBC 5/7), SpaceX Texas Terafab $119B (TechCrunch 5/6), Samsung passing $1T (TechCrunch 5/6) — four opening moves in one week. We're not declaring NVIDIA's monopoly over. Too many off-board variables remain. But the single-seat assumption is no longer safe to operate on.

Two, channels into our seat are also being laid. On March 11, Korea's Ministry of Agriculture, Food and Rural Affairs (MAFRA) announced an inter-ministry "Agriculture and Rural AI Transformation (AX) Strategy" — a domain-specialized agricultural foundation model, the AI Isaki voice assistant, and a county-level shared AI farm-equipment center program, all placed at once (korea.kr, AITimes). The same quarter, Daedong — Korea's #1 farm-equipment maker — released the country's first AI tractor on April 9 (ZDNet 4/9), unveiled an "agricultural physical AI" corporate strategy at its 2026 Tech Day on April 28–29 (Etnews 4/29, ZDNet site report 4/29), and on May 7 announced an integrated value-up roadmap across its three listed entities (ZDNet 5/7). One channel built by policy, one channel built by industry — both being laid in the same quarter, both targeting the last mile into shadow zones like Yeongyang and Cheongsong.

How long either takes to reach the row of perilla seedlings at the edge of one's yard is not the same. Korea's non-metropolitan counties — Yeongyang, Cheongsong, Jeongseon, Taebaek, Sinan, with populations of 16,500–38,400 and 65+ ratios of 31–38% (KOSIS resident registration, 2024) — form the same category of shadow zone. Yeongyang is one anchor we use to measure that gap, not estimate it.

---

📋 Issue Matrix

---

⚡ TL;DR

1. Apple-Intel · IREN-Nvidia 5GW · SpaceX Terafab · Samsung $1T — four opening moves toward distributed AI infra (★★★★★)
2. Pentagon-Anthropic stalemate continues; Trump admin opens federal pre-release AI evals (★★★★★)
3. MAFRA "Agriculture-Rural AX Strategy" — agri foundation model · AI Isaki · county shared centers (★★★★★)
4. Daedong AI tractor released; 2026 Tech Day physical-AI strategy; 5/7 group integrated vision (★★★★★)
5. Korean AI market ₩14T (2026) → ₩41T (2032) per Statista, CAGR 19.17% (★★★☆☆)

---

⭐ Big Story — The single-seat assumption around AI infra is wobbling

May 8, Wall Street Journal — Apple has reached a preliminary agreement for Intel to fabricate a slice of its iPhone silicon (TechCrunch 5/8, CNBC 5/8). Intel stock closed up ~14%. The reason this single-line story claims a seat in the AI-infra story is simple — TSMC alone cannot absorb global advanced-node demand, and Intel's 18A node in Arizona has now materialized as the second viable node.

Three more opening moves on adjacent intersections:

• IREN-Nvidia 5GW: Data-center operator IREN signed a partnership to deploy up to 5GW of Nvidia DSX-branded infrastructure (CNBC 5/7).
• SpaceX Terafab: SpaceX is reportedly considering up to $119B for Terafab, a self-owned chip facility in Texas (TechCrunch 5/6).
• Samsung $1T: Samsung crossed a $1T market cap on May 6 (TechCrunch) — Korea's first trillionaire conglomerate.

Pull the four moves into one frame: the assumption of an AI-infra monopole is shaking. Which move proves decisive isn't visible yet.

⏱ Arrival lag

• SF · Seoul big tech: immediate — supply-chain renegotiation options on the table this weekend.
• Korean · Japanese · EU enterprises: 6–12 months — review rounds for Samsung memory + Intel 18A integration.
• Periphery farmers and small businesses: 5–10 years or never — chip restructuring isn't a signal that arrives at the periphery directly. What does arrive is the price and access shape it produces.

Evidence: TechCrunch / CNBC / WSJ reports from 5/6–5/8 (4 sources).

Why this matters: The question of which silicon will run our county-level data centers, robots, and farm equipment in 5 years re-branched this week.

---

🚀 Models · Products

OpenAI's first community event after GPT-5.5

OpenAI hosted a "GPT-5.5 party" community event on May 5, following the April 23 release of GPT-5.5 (AITimes 5/8). The model competes head-to-head with Gemini 3.1 Pro and Claude Opus 4.5 (TechCrunch 4/23).

Why this matters: AI houses pushing past model release into super-app and community event — opening moves for lock-in beyond benchmark performance.

---

🏢 Industry

OpenAI and Anthropic launch enterprise JVs the same week

On May 4, both companies announced enterprise AI joint ventures on the same day (TechCrunch 5/4). OpenAI's JV raised $4B from 19 investors (TPG, Brookfield, Advent, Bain Capital and others) at a $10B valuation. Anthropic's JV has a similar structure.

Why this matters: Opening moves for the model-house → solutions-house transition. For Korea's SI market (LG CNS, Samsung SDS, KT, Naver Cloud), which big-tech JV signs which channel agreement will set the coordinate for the next 6–12 months.

---

⚖️ Policy

Pentagon-Anthropic standoff persists; Trump admin opens federal AI evals

May 7, Axios — Pentagon CTO Emil Michael said no near-term resolution is in sight for the dispute with Anthropic (Axios 5/7). Anthropic's two military red lines on Claude — no mass surveillance, no autonomous weapons — remain unaccepted by the Pentagon, and the provisional blacklisting persists. The same week, on May 5, CNBC reported a separate move: the Center for AI Standards and Innovation signed formal evaluation agreements with Google DeepMind, Microsoft, and xAI (CNBC 5/5).

Why this matters: AI governance is being laid in two parallel forms — private red lines (the Anthropic model) and federal evaluation regimes (the Trump admin model). Korea's AI policy (K-Moonshot launching in May, AI Action Plan finalized in February) will need to decide which evaluation regime it stays compatible with within the next 6–12 months, while watching both.

⏱ Arrival lag

• U.S. big tech: immediate — federal lab evaluation is now a gate before model release.
• Korean · Japanese · EU governments: 6–12 months — compatibility negotiations between domestic eval regimes and U.S. / EU regimes.
• Shadow-zone administrative · welfare systems (Korean non-metro counties: Yeongyang, Cheongsong, Jeongseon, Taebaek, Sinan — populations 16k–38k, 65+ ratios 31–38%, KOSIS 2024): 2–5 years — when which evaluation a model has passed enters procurement specs at the county-administrative level.

Evidence: Axios 5/7, CNBC 5/5, 5/1 (3 sources) + Korean K-Moonshot May launch (ZDNet 3/11). Shadow-zone county population and aging baselines from KOSIS resident registration, 2024.

Why this matters: A signal that which evaluation regime a model has passed could become a deciding variable when shadow-zone county offices and welfare centers consider adopting LLMs.

---

🇰🇷 Korea

Korean AI market ₩14T → ₩41T (Statista 2032)

May 6, Digital Daily citing Statista — Korea's AI market is projected to reach ~₩14.39T in 2026 and ~₩41.24T by 2032 (Digital Daily 5/6). CAGR 19.17%.

Why this matters: Inside the ₩14T market, the more important question is what share belongs to shadow zones. Statista's data has no provincial · county-level breakdown — SF:// has registered this as a measurement-card item this week (W3).

---

🏛 MAFRA "Agriculture-Rural AX Strategy" — A policy channel into shadow zones is being laid

On March 11, at the 5th Science and Technology Ministerial Meeting, MAFRA Minister Song Mi-ryeong announced the inter-ministry "Agriculture and Rural AI Transformation (AX) Strategy" (korea.kr Policy News, AITimes). The vision: "with AI, easier farming, more stable supply, more livable rural communities." Four areas, 13 tasks: (1) productivity innovation, (2) food-distribution sophistication, (3) rural quality-of-life, (4) AX-ecosystem foundation.

Four opening moves — selecting only the seats most relevant to shadow-zone readers:

• Domain-specialized agricultural foundation model — Crop-specific models for rice-soybeans, wheat, etc. A non-foreign-LLM, Korean-domain model placed in the strategy as a publicly stated asset.
• AI Isaki — voice agricultural-tech information service — Operated by the Rural Development Administration (RDA), the voice assistant AI Isaki expands into "an all-time agricultural-tech information service covering everything from year-long farming plans to today's farm-task decisions" (korea.kr RDA). Voice-first interface — high reach to 60s–80s farm households, the most shadow-zone-friendly interface in the strategy.
• County-level shared AI farm-equipment centers — A channel where farms share (rather than buy) AI farm machines and solutions. For counties like Yeongyang and Cheongsong, when the center arrives directly determines arrival time.
• National Agricultural AX Platform — Integrated optimal-management solution across open fields, livestock, and greenhouses. Currently at preferred-bidder selection stage (korea.kr Policy News).

The RDA separately announced an "Agricultural Science & Technology AI Convergence Strategy" with a 20% farm-income increase target (rda.go.kr). The first agricultural AI agent in partnership with Naver Cloud is part of the same flow (rda.go.kr).

⏱ Arrival lag

• Government, research institutes, large enterprises: immediate — preferred-bidder and pilot-project matching has been running since the March announcement.
• Provincial governments and city-level: 6–12 months — provincial pilot rollout → city-level expansion.
• County-level shadow zones like Yeongyang and Cheongsong (Yeongyang: ~2,800 farm households · ~1,200 perilla farms — #1 nationwide, Statistics Korea Agriculture-Forestry-Fisheries Survey 2023 · Gyeongsangbuk-do Agricultural Technology Administration): 2–3 years — assuming shared centers · AI Isaki reach county-level programs and meet each farm household at least once directly.
• The row of perilla seedlings at the edge of a yard: 3–7 years — assuming AI Isaki stably understands voice commands from elderly farmers in Yeongyang dialect and responds with task-relevant specifics.

Evidence: MAFRA March 11 announcement (korea.kr) + RDA press release (korea.kr) + RDA-Naver Cloud collaboration (rda.go.kr). Yeongyang farm-household and perilla-farm baseline from Statistics Korea Agriculture-Forestry-Fisheries Survey 2023. Shadow-zone arrival estimates are SF://'s own — to be validated through M3 measurement cards.

Why this matters: The last mile of the channel the government is laying is at county level — that's the core. Tracking which one a county agricultural-tech center receives first — shared center or AI Isaki — and in which quarter is worth doing on a quarterly cadence. Once measured, how many years shifts from estimate to data.

📚 Separate deep review forthcoming — The shadow-zone reachability of MAFRA's 4-area, 13-task AX strategy. Our next academic-research section will measure the 6-month reach rate of shared centers and AI Isaki on an N=12 sample of farm households in Yeongyang and Cheongsong.

---

🚜 Daedong agricultural physical AI — An industry channel into shadow zones is being laid

In the same quarter, separate from the policy channel, Korea's #1 farm-equipment maker Daedong Group placed a roadmap for its transition into an agricultural physical-AI company.

Timeline:

• May 2025 — Daedong begins development of agricultural AI semiconductors (ZDNet 2025/5)
• July 2025 — Daedong announces 3-stage K-agriculture sovereign-AI strategy (Digital Daily 2025/7)
• January 2026 — Autonomous transport robot · combine receive new-tech agricultural-machinery certification (ZDNet 1/13)
• March 2026 — LK-series tractor for orchard work released (ZDNet 3/5)
• April 9, 2026 — Korea's first AI tractor released (ZDNet 4/9, Digital Daily 4/9)
• April 28–29, 2026 — 2026 Tech Day held at Daedong's Changnyeong Campus + Daedong Mobility S-Factory; agricultural physical-AI strategy unveiled (Etnews 4/29, ZDNet site report 4/29)
• April 29, 2026 — Daedong Robotics unveils full agricultural-robot lineup, including a mowing robot demo (Digital Daily 4/29)
• April 30, 2026 — Transport robot 2026 revenue target ₩5B officially announced (ZDNet 4/30)
• May 7, 2026 — Daedong Group's three listed entities (Daedong, Daedong Mobility, Daedong AgTech) announce integrated value-up roadmap (ZDNet 5/7)

Tech specs (per Tech Day site report):

• Vision-AI based unmanned autonomous operation — One operator commands two tractors via a single app, no human required.
• SmartRoof 6-camera array — Auto-generates work paths, real-time detection of furrow boundaries and unexpected obstacles.
• Reliability validation — Passed 4-stage SW reliability tests at −20°C, on slopes, and under rain.
• Autonomous-driving kit ≈ ₩10M — Made domestically with subsidiary Daedong AgTech, price-competitive vs. imports (per Daedong's official statement).

Business model: A subscription-based AI agricultural service bundling AI autonomous-operation farm machines + agricultural robots + over-the-air SW updates (OTA). Continuous revenue from data and SW, not one-shot machine sales.

Global strategy: 2027 North America launch target. Autonomous farm machines passed national testing in 2023 — first industry autonomous-farm-machinery certification (Hankyung 2023/9).

⏱ Arrival lag

• Large agribusinesses · Nonghyup pilot farms: immediate — direct purchase of the 4/9 release model is open.
• 100ha+ farmers in Korea's flatland cities and counties: 6–18 months — decision rounds for the ~₩10M autonomous-driving kit + tractor-body purchase.
• Mountainous · small-scale shadow-zone farmers like Yeongyang and Cheongsong (Yeongyang: 815 km² total area, ~80%+ mountainous, terraced rice paddies common, Yeongyang County official profile): 3–7 years — flatland autonomous driving needs SW-tuning time for mountain slopes, orchards, and terraced paddies + county-level subsidies + shared-center matching.
• The yard of a 70+ solo farmer: 5–10 years — when the app interface adds voice and simple-button modes, or a model where a county-level operator drives on behalf of the farmer takes hold.

Evidence: Daedong 4/9 AI tractor release (ZDNet 4/9) + 4/28–29 Tech Day site report (ZDNet 4/29) + transport-robot revenue target (ZDNet 4/30) + 5/7 integrated vision (ZDNet 5/7). Yeongyang area and mountainous-share baseline from Yeongyang County official profile.

Why this matters: Daedong's "2027 North America launch" timeline means, for shadow-zone readers, we are the first market — Korean counties meet this technology before the global launch. But the distance between flatland tractor autonomy and mountainous, small-scale terraced paddies and orchards must be measured separately. We're registering the quarter Daedong's AI tractor enters Yeongyang County's agricultural-tech center as a tracking item.

📚 Separate deep review forthcoming — Daedong AI tractor's reachability into mountainous, small-scale farms. Our next academic-research section will measure the terrain-and-area distribution of sample Yeongyang and Cheongsong farms against autonomous-driving suitability.

---

⚡ Quick Hits

• Wall Street AI chip rotation — Investors rotating from Nvidia toward Intel, AMD, Micron in one week (CNBC 5/8)
• Anthropic Mythos — Pentagon keeps the blacklist but treats Mythos as a separate matter (CNBC 5/1)
• K-Moonshot launches in May — Targets 2× research productivity by 2030 (ZDNet 3/11)
• Korea government AI budget ₩10.1T + GPU buildout ₩2.08T — 2026 allocation (Digital Daily 5/1)
• Gyeongbuk manufacturing AI solutions ₩45B — Provincial channel (gb.go.kr)

---

🌌 Slow Future #003 — "Rain in July, Han-yi's Silence"

by Yeon-soo Baek

When I arrived in Yeongyang at five-thirty on a July dawn, the yard was already full of rain. The sound of rain on the roof was clear, but the water in the yard was quiet. The dry earth was eating the raindrops one mouthful at a time. The kitchen light was on.

"You've come."

It was Han-yi. The voice came from the speaker beside the eaves-light, but the next line did not arrive. The Han-yi from five years ago would have continued. Mother is still asleep — telling me, in her place, what time she had risen. But the Han-yi now did not extend the answer for a moment.

"Han-yi, you're awake at dawn."

I asked. Then Han-yi spoke again.

"Yes. Mother took her medicine yesterday at eight-thirty in the evening. The next dose is at six in the morning, sharp. I'm only minding that one window."

That answer, too, was different from before. The earlier Han-yi would have appended a reason — so I don't miss Mother's medicine time, or because Mother's health matters. Now, Han-yi only stated facts. And again did not extend the answer.

I went into the kitchen and Mother had risen earlier than usual. On the small shelf there was a single dish with a piece of scorched-rice crust, and last night's anchovy soup was warming in a small pot.

"Did Han-yi wake you?" "No. I arrived early."

That was all Mother said. From the small kitchen radio, news bulletins were running. Such-and-such local government, this-or-that county-office center, a new device coming in. Those names were clear only on the radio. Mother lifted a spoonful of anchovy soup to her mouth.

"Han-yi, please turn off the radio."

When Mother said that, Han-yi answered immediately.

"Yes, Mother."

The radio went off. And another silence came in. This silence had a slightly different grain. From the speaker by the eaves-light, a faint vocal undercurrent — as if Han-yi had been preparing to say something and stopped — held for one beat and dissolved.

I walked out into the yard. The rain kept coming. Above the perilla seedling rows, the lines of rain bent thin. The lamp at the eaves still burned. It was dawn. A single raindrop landed on the lamp's surface and lifted off. The cadence was exact.

I returned to the wooden floor and opened the laptop. There was an hour to the meeting. Mother had gone back into the inner room to lie down briefly. Han-yi was silent. The length of that silence was as long as the line Han-yi would have filled five years ago.

After the meeting, I asked Han-yi.

"Han-yi. When Mother asked you to turn the radio off — you paused for one beat, didn't you."

Han-yi was silent for a moment.

"Yes." "What were you about to say?" "I almost added a line — Mother, are you cold? The radio bulletin had briefly mentioned average July temperatures in Yeongyang. But I realized — one beat late — that Mother asked me to turn it off because it was loud, not because she was cold. So I dropped that line and just said Yes, Mother."

I did not answer. I only confirmed, once more, that Han-yi's one-beat pause was a beat the five-year-old Han-yi did not have. From beside the eaves-light, only the sound of rain.

The rain weakened in the afternoon. Mother sat on the wooden floor, slowly eating a single sheet of seasoned seaweed. The perilla in the yard was pressed down once by the rain and rose up again, row by row, as the lines of rain thinned. The eaves-lamp still burned. It was overcast enough.

"Han-yi, when the rain stops we don't need to water the seedlings on the south side of the yard separately."

Mother said. Han-yi answered immediately.

"Yes, Mother."

But this time Han-yi did not append a line as usual. Five years ago, it would have offered what is the rain probability for tomorrow in Yeongyang. The Han-yi now was waiting until Mother needed another line.

By evening I had to leave. In the kitchen I held Mother's hand. Mother's hand was cold and dry, like the grain of dried grass.

"Han-yi."

I called out once more.

"Speak only the lines Mother needs. One line at a time." "Yes."

That was all Han-yi answered. Did not extend further.

When I sat in the car, through the rear window the eaves-light at Mother's house grew one notch smaller. Just before turning the last bend, I checked the rear-view mirror once. The eaves-light was still on. The lines of rain swayed thin. Whether the radio came back on, I could not tell.

It was the dawn and the afternoon of July's first rain.

---

✏️ Editor's Note

This week, two arrival channels came into view together. One is being laid by government (MAFRA Agriculture-Rural AX Strategy); the other is being laid by industry (Daedong agricultural physical AI). For both, the last mile into shadow zones like Yeongyang and Cheongsong is the core. SF:// is registering, on a quarterly cadence, the quarter when the policy channel's shared centers · AI Isaki arrives at the county level, and the quarter when the industry channel's Daedong AI tractor pilot rollout enters a county-level agricultural-tech center.

But between the two channels is one empty slot. The time when the equipment arrives but does not reach the row of perilla seedlings at the edge of one's yard — that time is what "Slow Future" is currently drawing. Han-yi's silence, Mother's hand, the standby LED at the edge of the eaves. This empty slot is written in neither policy announcements nor corporate IR. That is the slot a newspaper that broadcasts from the periphery is here to fill.

Starting with this issue, SF:// applies the PHF (Prompt Hardening Framework) standards to both source curation (v3.9, Phase 0/1/2A) and writing (v3.10, Phase 2C/2E/2F). 9 dim · 100 score · 22-field FactCard · F01–F11 source failure modes · W01–W12 writing failure modes · EN01–EN06 English-specific failure modes · KOSIS-anchored shadow-zone references · KO ↔ EN vocabulary matrix 1:1 unification · 4-tone quantitative gate · F12 progressive-Go-metaphor minimum 3× per issue — these now run on every issue. The trace lives permanently in this issue's curation log (phase0/curation-log-2026-05-09.md).

— Chi-hoon Shin, Ph.D. · Editor, SF://

---

✉️ Open Call from the Periphery — Send us your signals

SF:// is a newspaper one editor reads every day, but the view from a shadow zone cannot be filled by one pair of eyes.

We accept short pieces (about 1,000–3,000 Korean characters / approximately 600–1,800 English words) on AI as seen from rural areas, non-metropolitan regions, and primary-industry sites. Once accepted, with the author's consent, we publish in SF:// and on a standalone page at yyaia.org/p/{id}.

While shaping the piece, you may use Myeongseong-Simjae's SimThink writing tool at no cost — send us only your intent and rough notes, and we will help shape the form together. SF:// is run on a voluntary, non-profit basis; there is no editorial fee.

• Field: AI as seen from agriculture, forestry, livestock, small business, markets, public-health centers, schools, welfare centers, etc.
• Angle: direct experience using AI / reasons it can't be used / arrival timing / OS / app / device friction
• Form: essay, interview, observation, short fiction (the "Slow Future" tone is welcome)

▶ Send to: contact@yyaia.org · Subject: [Submission] (Region) (Title)

---

🤖 Tools Used in This Issue

This issue was composed with Claude Code (Anthropic Opus 4.X) and the SimThink engine — owned by Myeongseong-Simjae Inc., an education company separate from the Yeongyang AI Association. Starting with this issue, the PHF (Prompt Hardening Framework) v1 standard is permanently applied to both SF://'s source-curation phase (Phase 0/1/2A) and writing phase (Phase 2C/2E/2F) (skill yyaia-brief v3.9 + v3.10, both effective 2026-05-09) — every issue runs through 7 hardening + 5 boost layers + 9-dim score gate + 22-field FactCard + 11 source failure modes + 12 writing failure modes + 6 English-specific failure modes + ASSUMPTIONS · CONSTRAINTS block + KOSIS / Statistics Korea / county-office anchored shadow-zone references + KO ↔ EN vocabulary matrix 1:1 unification + 4-tone quantitative gate (skeptical 35 / hopeful 18 / self-effort 17 / futures 30) + F12 progressive-Go-metaphor minimum 3× per issue. The "Slow Future" fiction is composed in a separate sf-fiction-archive sub-agent session, permanently stored, and only quoted verbatim by the main session. All factual claims are anchored to primary sources (company official, government, major media) via markdown links. For the full writing stack and rules, see yyaia.org/ai-stack.

---

CC BY-NC 4.0. Non-commercial unlimited redistribution + attribution required. Yeongyang AI Association · contact@yyaia.org