본문 바로가기
IT Strategy Weekly

Oct 25 – Oct 31, 2026

Korean original

Not investment advice — these are resources for learning and forming your own view.

The thread running through this week’s technology news is that AI’s bottleneck has moved from the model to complementary resources such as power, data and trust. The question for leaders is therefore not which model to use, but which resource you hold that others cannot get.

The strategist's view

The moat in AI competition has already moved from the model to electricity and contracts. Write down now what your company holds that cannot be copied.

Taken together, this week's news points in one direction. Microsoft reported quarterly Azure revenue of $33.2 billion and said OpenAI alone accounted for $10 billion. In the same week, Amazon posted negative free cash flow of $2.4 billion even though its cloud business sold well, and Anthropic, ahead of a listing, is being discussed for a speculative-grade credit rating. Money is flowing in at one end and out at the other. The dividing line is not model performance. It is who controls capital, power and distribution channels.

The list of 'hard-to-imitate resources' in the resource-based view has been rewritten over the past two years. It used to be parameter counts and benchmark scores. Now it is grid-connection approvals that take years, real-world data scarce enough that companies buy the codebases of failed startups, and sales channels that reach inside large enterprises' procurement organizations. A model can be caught up with in months. These three cannot be caught up with by spending money alone, so once a gap opens it rarely closes.

So in executive meetings, do not ask about the 'AI adoption rate'. Ask three things. What percentage of the power you will need in three years have you already secured? Which of the data generated only by your operations do competitors lack? And whose channel, already trusted by your customers, will your AI product ride in on? If you cannot answer these three in numbers, then however good the model you attach, the financial return is likely to converge on zero, as Apollo warned.

Cases through a framework 3

Resource-Based View (RBV)

The data center boom has an upside: innovation for the US grid

As AI data center power demand runs into the limits of existing grids, power infrastructure such as grid modernization and solid-state transformers has emerged as a new growth market. Boom Supersonic, which was building supersonic airliners, has shifted its business toward a factory for gas turbines for data centers. Intel's shares plunged under AI-driven component supply pressure, wiping $12 billion off the value of the government's stake alone. Energy, not compute, has become the gatekeeper of growth.

Power connection capacity, substation equipment, sites and grid-interconnection permits cannot be bought on the market just because you have money. They come only after queues lasting years and local regulatory approval. Holdings therefore differ widely between companies, and the gaps are not easily closed, which is exactly the premise of the resource-based view. Anyone with money can buy the same GPUs, but not the same power contracts and sites. That asymmetry is the common background of this week's stories.

This news is often read as a thematic-stock story: AI raises power demand, so utilities benefit. Through the resource-based view, however, the point is not the rise in demand. It is that the ranking of inimitable resources has changed. Model performance can be matched in months, but matching secured power capacity takes years. The real moat in AI competition is moving from the algorithm to the electricity contract.

Decision prompt — When drawing up AI investment plans, create a separate 'power procurement roadmap' document, apart from the GPU procurement schedule. Put numbers on the gap between the power needed over the next three years and what is already secured. Within this quarter, decide how to fill the shortfall: on-site generation, long-term PPAs, or moving workloads to regions with spare power. If your company has manufacturing and engineering capability, also put on the due-diligence list whether it could enter power equipment supply, as Boom has.

About the framework

Why do some firms in the same industry keep outperforming? This view looks for the answer inside the firm rather than in its market position: resources and capabilities that others do not have. Not just any resource qualifies. It must be valuable, rare, inimitable and non-substitutable (VRIN). Patents, data, proprietary algorithms and talent, which cannot be bought for money, fall into this category.

TechCrunch

Crossing the Chasm

AI delivers ‘zero financial payoff’ to most companies, Apollo economist says

Apollo's economist warned that AI has not produced measurable financial results at most companies outside the tech sector. In the same week came an 'AI graveyard' article collecting failed AI projects, and an analysis arguing that ROI fails to appear because companies cannot connect their own context to the model. Ollie, a privacy-focused AI assistant, shut down. Meanwhile a Databricks co-founder said the key to winning large enterprise customers is not technical excellence but the existing sales channels of giant companies.

AI adoption is not a continuous improvement layered onto existing workflows. It is a discontinuous technology that asks the people doing the work to change how they work. The innovation teams that embraced it early were drawn by novelty. The pragmatic buyers now at the approval stage demand references from peers in their own industry and complete solutions. This mismatch in demand is showing now, as reports of missing financial returns and lists of failed projects arrive together.

This news is often read as a sign that the AI bubble is deflating. Through the lens of the chasm, what is deflating is not the technology but demand from the early-adopter segment. What is needed now is not withdrawing investment but narrowing the scope to one industry and one task, and building a 'proven complete solution' and references. That is also why Databricks' advice matters. The bridge across the chasm is not a better model but a channel that mainstream buyers already trust.

Decision prompt — Re-review all in-house AI pilots in full. Keep only those where cost savings or revenue gains can be tracked as income statement line items within six months, and stop the rest. If you are a supplier, drop generic positioning and narrow to a complete package that takes full responsibility for one task in one vertical. Sign, within this year, a partnership that places it on the marketplace or sales channel of a large vendor that customers in that industry already use.

About the framework

On the path where a new technology moves from sales to a small group of enthusiastic early adopters to the mass market, there is a deep chasm. Early adopters are drawn by the novelty of the technology itself, but pragmatic mainstream customers want proven complete solutions and references, so what the two groups want is fundamentally mismatched. Early success therefore does not guarantee mass-market success, and many technologies lose funding and momentum in this chasm.

TechCrunch

Porter's Five Forces

Anthropic VS code extension on Cursor backdoored for nearly 12 hours

Anthropic's official Claude VS Code extension was distributed on the Cursor marketplace with a backdoor for about 12 hours, exposing the systems of many developers to risk. In the same week, Applebee's leaked customer data through a database password hard-coded in its app, Binance halted withdrawals amid suspicion of a large-scale hack, and Bank of America told customers not to hand their bank passwords to AI agents, shifting responsibility for incidents onto them.

The AI development tools market has clear boundaries as an industry, with distinct model providers, editor and marketplace operators, extension developers and enterprise buyers. This incident shows starkly how supplier bargaining power works inside it. If a marketplace passes along a binary signed by the model company and enterprises install it without verification, then when one link in the supply chain is breached, the profits and trust of the whole industry thin out together.

This news is often read as 'another supply chain hack'. Through the lens of the five forces, it is a structural problem, not an accident. Enterprise buyers' bargaining power is close to zero, so they cannot take on the burden of verification. Suppliers such as Bank of America are making clauses that push responsibility onto users the standard. Unless contract terms change now, the more AI tools buyers use, the more risk accumulates on the buyer's side.

Decision prompt — Move AI coding extensions and agents to an internal allowlist so that only items that have passed signature verification and version pinning can be installed, and replace all credentials on developer machines with short-lived tokens. At the same time, reopen the incident liability clauses in every AI tool contract, and this quarter draw up a list of contracts to renegotiate so that suppliers specify breach-notification deadlines and loss sharing.

About the framework

How profitable an industry is depends not on how capable its firms are, but on the structure of the industry. Porter summarized that structure as five forces: existing rivalry, the threat of new entrants, substitutes, supplier bargaining power and buyer bargaining power. The stronger these forces, the thinner the profits of everyone in the industry, so one first reads these five axes to decide which industry to enter and where to position within it.

Ars Technica

Based on 114 items over 7 days

Past issues