大型机优势：AI时代关键业务系统的负载部署与运行决策

Workload placement andexecution decisions formission‑critical systems Contents IntroductionMainframe makes a comeback..................................................................5 Part oneIntelligent execution: revisiting cloud-first assumptions...............................7 Part twoProduction economics: governing cost in operational execution..................10 Part three AI trust: No trade-offs allowed................................................................12 Action guide Five moves leaders must make now to anchor Foreword As AI scales into production, enterprises are re-evaluating where their most criticalworkloads should run—including which workloads are best suited for the mainframeand which should go on public cloud and distributed platforms. Rising cost volatility, tighteningregulatory and sovereignty expectations, and escalating resilience requirements—now magnified For leaders, the question is no longer“Where can we run AI?”, but“How do we bring AI to theapplications and data where mission‑critical execution occurs?” This brief gives CIOs, CTOs, enterprise architects, and infrastructure leaders clear, data-backedevidence on where mission critical workloads are being placed, why those decisions are changing, What every technology leader needs to know—now –AI raises the stakes: when critical work runs, failure is no longer acceptable.–Enterprises are shifting their thinking from cloud-first to execution-first.–Mainframes matter most where the business can’t afford delay, disruption, or doubt. Key takeaways Hybrid now dominatesenterprise IT, with themainframe anchoring Cost overruns arevisible, but costgovernance is not. Cloud economics haveunderperformed; AI delivers more valuewhen decisions happen Despite rising cloud overruns, only14% consistently compare costsacross platforms. As AI increasesongoing compute demand, weakcost discipline becomes a serious As AI moves into live transactions,executives want it closer tosystems of record. Only 21%run AI inference with systems 74% of executives are movingaway from defaulting to cloud-firstapplications, instead selecting theplatforms best suited forworkflows. This shift signals that 72% of organizations reporthigher-than-expected cloudproduction costs. By contrast,executives rate the mainframe Introduction Mainframe makes New research from the IBM Institute for Business Value (IBV), conducted incollaboration with Oxford Economics, shows that a dramatic shift is underway.Based on a global survey of 425 senior technology leaders—each directly For over 60 years, mainframes have quietly built an unrivaled reputation forreliability and availability. Today, organizations can achieve up to 99.999999%availability, which in practical terms, translates to less than one-third of Even so, cloud has dominated enterprise IT for the past decade. But AI isforcing a rethink of this cloud-first approach. When AI is embedded directly intothe transaction path, the limits of cloud-centric architectures surface quickly: Respondents represented organizations across multiple industries—includingbanking, insurance, manufacturing, consumer and retail, healthcare and lifesciences, energy and utilities, telecommunications, and government—and Customers don’t care where transactions happen—they just care that theywork. They don’t want excuses, they want reliability. None of this is an IT detail,it’s a business imperative, and it’s built one transaction at a time. Nearly three-quarters of executives surveyed are moving away from cloud-firstdefaults. This shift may be partly a response to higher-than-expected cloudproduction costs. 72% report surprise cloud expenses, and respondents rate the And if the transactions are an organization’s most precious assets, then today’smainframe plays an important role in enabling real-time insight, supporting This isn’t a return to old technology, but a renewedrecognition of where execution matters most. Asorganizations re-anchor mission-critical workloads This report explores an architecture decision frameworkthat we call an outcomes triangle (see Figure 1).We’ll examine each element to better understand First, we’ll dig into what’s required to deliver intelligentexecution at scale, and why existing assumptions needto be revisited when AI comes into play. Second, we’ll look Part one Intelligent execution:revisiting cloud-first AI changes the architecture conversation becauseit changes the cost of failure. What is emerging now When intelligence sits at the edge, slowdowns might be tolerable. But when AImoves into the transaction path, they are not. In this context, the mainframeisn’t a box sitting outside the workflow. Rather, it can connect and host modernenvironments—including private clouds—by running Linux and containerizedworkloads and by exposing APIs and event streams, so it becomes fully of executives report shiftingtoward outcome-basedplacement for their most Figure 2Where AI inference