Why does the market increasingly question risk, data, and cash flow as AI booms?

4. Silver Lake in acquisition talks with Workday: AI may not rapidly eliminate traditional enterprise software After Reuters exclusively reported that Silver Lake was in acquisition negotiations with Workday, Workday’s stock surged nearly 18%, pushing its market capitalization above $51 billion—though the deal has not yet been formally finalized. Traditional enterprise software faced valuation compression this year amid concerns that “AI can write code and generate applications.” Yet analysts point out that migration costs for core systems—such as HR, finance, and CRM—are extremely high; recent strong financial results from companies like ServiceNow further support the view that AI is more likely to augment, rather than instantly replace, these foundational systems. The proposition that “AI will kill software” overlooks a critical variable: software’s value stems not only from surface-level functionality but also from long-accumulated data, permissions, historical records, employee habits, and accountability chains. While AI can rapidly replicate a given function, it cannot easily replicate, in short order, the deep embeddedness of a system within an organization. For enterprise customers, replacing a core system entails migration risk, compliance risk, and operational disruption—meaning that “functional substitutability” does not automatically equate to “full-system substitutability.” A more stable underlying pattern is this: the deeper a tool is embedded into workflows, the higher its replacement cost. When building AI products, it is often more effective to embed AI into existing tools—Excel, security logs, Obsidian, or video production pipelines—than to build an isolated chat interface. The most commercially viable and verifiable opportunities frequently arise from reducing migration friction while preserving existing accountability structures—not by demanding that customers tear everything down and start over. What deserves greater validation is how AI can reduce friction in legacy systems—not whether it can demonstrate a flashy new UI. 5. RWE nearing data center agreements: “Land with power” emerges as a new AI asset German energy giant RWE stated it is nearing agreement with data center operators on two projects. The company owns approximately 30 sites equipped with robust electrical infrastructure, about 10 of which are deemed suitable for data center development. Management explicitly noted that land and grid connections previously earmarked for power plants, battery storage, and similar projects can now be directly repurposed for AI data centers. This information comes from RWE and a Reuters report dated August 13. Why have energy companies’ legacy sites suddenly become more valuable? As data centers scale up, site value is no longer determined solely by land area—it hinges equally on grid capacity, substation infrastructure, permitting status, fiber connectivity, and cooling conditions. Surging compute demand has pushed grid interconnection to the top of the bottleneck list—and transformed traditional energy firms’ underutilized sites into newly strategic assets. As the digital industry penetrates deeper into the physical world, its expansion becomes increasingly constrained by construction timelines, regulatory approvals, and finite physical resources—not just chip orders. This reveals a broader principle: when industrial bottlenecks shift, legacy assets that control the *new* bottleneck get revalued. Likewise, identifying verifiable industrial AI opportunities requires auditing enterprises’ existing—but underused—resources: historical incident records, surveillance video archives, hazard logs, expert knowledge, and equipment operational data. Organizing these assets into a layer of data that is callable, licensable, and verifiable may prove far more valuable than building a brand-new user-facing interface from scratch.

The true cost of AI data centers: beyond GPUs, it’s the grid, water, and noise that count.

AI data centers aren’t just about buying GPUs—the real costs lie in grid capacity, water resources, noise, and contract transparency. Original source: AI Xiao Lao Liu (AI Little Rogue) The AI industry loves talking about GPUs, models, tokens, and inference costs—but tends to avoid discussing the grid, water, noise, diesel generators, and residential electricity bills. Once a data center is physically sited, abstract “compute demand” transforms into tangible, concrete realities: hundreds of megawatts of load, cooling systems, transmission capacity, backup power, land use—and a contract invisible to outsiders. Figure: AI data centers convert abstract compute demand into real-world pressure on the grid, water supplies, and community space. First, consider several key figures: | Project | Figure | Meaning | |---------|--------|---------| | Firmus planned load | 400–500 MW | Total demand across two to three AI factory sites | | Bell Bay Aluminium load | 335 MW | Current largest single electricity consumer in the region | | Launceston single-site contracted load | 104 MW | Roughly equivalent to Boyer Paper Mill’s current electricity consumption | | 2024–25 net imported electricity | 1,550 GWh | Tasmania is not reliably power-surplus | | Launceston estimated permanent jobs | ~30 | Very few staff required for day-to-day operations | | Bell Bay estimated jobs | ~70 daytime, ~25 each for evening/night shifts | Employment density remains low relative to power consumption | Taken together, these numbers shift the question away from “Should we adopt AI?” A more precise question emerges: In a region where electricity supply is already tight—and existing industries are still waiting in line—is it worthwhile to allocate a large chunk of grid capacity to a foreign-owned data center? Data centers have very low employment density. Construction phases generate temporary construction jobs, and promotional materials often highlight large investment figures. But long-term operations rely heavily on automation, minimal on-site staffing, and specialized maintenance. A multi-hundred-megawatt campus may host only around 100 permanent staff. That differs sharply from traditional industry. Aluminum smelters and paper mills stimulate local supply chains—procurement, repair, logistics, materials, and services—all generating significant economic activity nearby. Data centers operate more like sealed boxes: servers, cooling, power, and networking; core equipment is imported; computing output is sold to distant customers; and the local community is left primarily with electricity strain and a handful of jobs. If a project consumes nearly as much electricity as two major industrial enterprises combined—yet delivers only a fraction of their employment impact—it cannot be justified solely by the label “high-tech investment.” “Green compute” must undergo marginal electricity audits. Assessing whether a data center is truly green cannot rely solely on the average generation mix of the local power grid. What matters critically is *marginal electricity*: where does the *next kilowatt-hour* come from when new load is added? If newly built wind, solar, or pumped-hydro capacity hasn’t yet come online—and the data center has already begun drawing power—the shortfall will be met by external grid imports and fossil-fueled generation. In such cases, marketing that compute as “clean AI” demands rigorous evidence: ・ Independently audited PUE; ・ Authentically traceable renewable energy matching; ・ Explicit load-flexibility clauses; ・ Mechanisms to prioritize electricity for residents and incumbent industries during peak demand.

Data Center Boom Raises Issue of Who Pays for Grid, RWE Says

17 hours ago

RWE CEO Markus Krebber states data centers driving electricity demand should fund grid expansion instead of spreading costs to all consumers. Google's premium power agreement illustrates growing tensions over infrastructure funding for AI growth in Europe.

bloomberg

2026-10-07

Data Center Boom Raises Issue of Who Pays for Grid, RWE Says

Data Center Boom Raises Issue of Who Pays for Grid, RWE Says

FAQ