Factory Data Is Not a Free Lunch
A factory dataset is worth precisely nothing if its owner cannot control it. That was the hard-nosed thread running through the ASI City Forum at Stanford on November 17: industrial AI is not just about smarter models or splashy chip announcements. It is about whether manufacturers will trust a system with proprietary operating data, whether the technology fixes an expensive problem, and whether somebody eventually signs a contract.
Jimmy H. Rhee, chairman of U.S.-based ASI Foundry and a former Maryland cabinet secretary, put the commercial point plainly. Korea has high-quality manufacturing data; the U.S. has advanced AI models. Marry the two while preserving local control over data, access and use, and there may be something real there. Ignore governance, and the whole thing turns into another conference-room romance that dies before procurement gets involved.
That distinction matters. Investors have heard enough broad AI chatter to fill a stadium. Industrial deployments live or die in grimy particulars: who owns the data, what system connects to what machine, how quickly a plant manager sees value, and who carries the blame when an output is wrong. Rhee's insistence on paying customers and contracts was the most useful line of the day. No revenue, no victory lap.
Chips Remain the Load-Bearing Wall
John L. Hennessy, Alphabet board chairman and Stanford president emeritus, described next-generation semiconductors as essential to continued AI progress. He is right to emphasize that point. Software excitement gets the headlines, but compute capacity, memory, power efficiency and dependable supply chains decide how far AI can actually reach beyond the data center. Alphabet's GOOGL board chairman is hardly a disinterested bystander in the wider AI race, yet the observation itself is hard to argue with.
Korea enters this discussion with genuine industrial muscle in chips, consumer products and automobiles. Labib Tazwar Rahman, president of Voltai and a Stanford adjunct lecturer, argued for joint development that pairs Korean manufacturing know-how with U.S. software and chip-design expertise. He sees a route for Korea to build leadership in next-generation chips and develop a new crop of semiconductor leaders on the foundation laid by Samsung and SK hynix.
Fair enough, but “next-generation” can become a dangerously elastic phrase. The semiconductor business punishes vague ambition with brutal capital bills and short product cycles. Cooperation is promising when it produces better yields, faster design iteration, secure capacity or a customer advantage. It is fluff when it amounts to ceremonial handshakes around a giant screen.
Trust Has to Be Engineered, Not Promised
Hennessy also stressed safeguards against AI harms and retraining for workers whose jobs will change. That deserves more than a polite nod. Industrial AI that raises throughput while leaving workers, communities and regulators to absorb the disruption is inviting a backlash. Governments and businesses both have a role in preparing people for new work, especially when the technology moves from pilot programs into broad production.
There is a practical business angle here, not merely a moral one. A company that cannot explain access controls, audit trails, workforce plans and accountability will find its sales cycle getting longer. Big manufacturers do not casually hand over sensitive production information, and they should not. ASI Foundry's stated focus on governance and local control speaks directly to that bottleneck. The proof, naturally, will be in actual deployments rather than forum language.
Megacities Need More Than Net-Zero Slogans
The later discussion on Korean megacity and AI infrastructure projects was refreshingly less starry-eyed. Stanford professor Michael Lepech argued that life-cycle assessment should guide decisions about resource demands and environmental effects. His point was sharper than the usual “be sustainable” wallpaper: energy and water use must be evaluated locally, while projects should deliver measurable environmental, social and economic benefits to surrounding communities.
That is the right frame. A data-intensive industrial ecosystem can look wonderful in a national presentation and still strain local grids, water systems or housing if planners wave away the arithmetic. Net-zero targets alone do not settle those tradeoffs. Where does the electricity come from? What does the water draw do in that specific place? Who benefits when the project succeeds, and who gets stuck with the inconvenience? Those are not side questions. They are the project.
- Industrial AI needs data governance that manufacturers can live with.
- Chip collaboration needs concrete technical and commercial milestones.
- Megacity infrastructure needs local resource accounting, not broad promises.
- Worker retraining and community outcomes belong in the operating plan.
Rhee added that durable industrial ecosystems also require supportive tax and regulatory structures, infrastructure, competitive companies, capital and accessible markets. That sounds obvious because it is obvious, yet regions routinely try to skip one or two pieces and act shocked when investments stall. The Stanford forum did not announce a transaction or solve those execution gaps. What it did offer was a sane checklist. In this corner of AI, the winners will not be the loudest evangelists. They will be the groups that can turn trusted data, chips, infrastructure and local legitimacy into work that customers will pay for.