2026年AI基础设施状况：系统能否承受AI规模化压力？

Can Systems Withstand AI Scale? Executive summary What 1,000+ senior cloud architects and technologyexecutives around the world are saying about theirorganization’s preparedness for AI workloads, and theirstrategies for the next year 2025 may be remembered as the year of outages and the year AI hit production at scale. But there is a link between these twophenomena and their relationship needs to be investigated. TheState of AI Infrastructure 2026report explores a critical tension shaping today’s enterprise architecture: the rapid expansionof AI workloads is outpacing the systems meant to support them. Based on a global survey of 1,125 senior cloud, engineering,and infrastructure leaders, the report reveals exactly how invested companies are in AI, the hidden cost of AI success and whattop-performing teams are doing to survive and scale. 4. The database layer is emergingas a critical point of failure. 1. AI growth is no longer optional–it’s guaranteed. 100% of respondents expect AI workloads to grow in thenext year. More than 60% predict increases of 20% ormore. At this point, AI adoption is an inevitability. The onlyquestion is how existing systems will respond to the volumeand velocity of what’s coming next. The data reveals anunsettling answer. 30% of respondents identified the database asthe first point of failure in an AI-overload scenario,second only to the cloud infrastructure itself. Theproblem is not running on the cloud, the problem ishow design decisions are made with regard to howdata is ingested, processed, stored, and moved. 2. Infrastructure failure is expected very soon. 5. AI is expected to drive ameaningful share of outages. AI is shifting from an innovation initiative to asystems-level risk. 83% of leaders believe their datainfrastructure will fail without major upgrades in thenext 24 months. The fragility of current infrastructurewon’t be solved with routine maintenance, it requiresan overhaul of how organizations approach thearchitectural foundation on which their systems rely. AI-related reliability issues are no longerhypothetical. 77% expect AI to drive at least 10%of all service disruptions in the year ahead. 6. The financial impact of AI-related downtimeis already substantial and rising. 3. For one-third of companies, the breakingpoint is less than a year away. As AI systems become embedded in core businessoperations, outages carry immediate and materialfinancial consequences. 98% of companies say an hourof AI-related downtime would cost at least $10,000;nearly two-thirds say it would cost over $100,000. 34% expect their infrastructure to fail within the next11 months. For a significant share of enterprises,infrastructure failure related to AI scale is viewedas an imminent event, not a distant possibility. real-time inference, and always-on automation areoverwhelming brittle backends across every industry. As AIadoption continues to accelerate, the winners will be thosewhose infrastructure can handle unprecedented scale.Organizations that can’t absorb constant, compoundingdemand will encounter outages, cost volatility, anddegraded performance long before AI ambitions arerealized. The path forward starts with distributed,resilient systems built to withstand continuous scale andoperate through success, not just recover from failure. 7. Leadership misalignment is accelerating the risk. Nearly two-thirds of respondents (63%) say their leadershipteams underestimate how quickly AI demands will outpaceexisting data infrastructure. This suggests that whilecompanies have been investing in AI, the investments havebeen too reactive, and may not truly prevent disaster. Today, resilience is not just a best practice, it’s thebattleground for the next phase of AI adoption. Theinfrastructure powering today’s enterprises wasn’t builtfor AI-native scale, and it’s showing. Recursive agents, Survey Methodology The State of AI Scale & Resilience Survey was conducted by Cockroach Labs andWakefield Researchamong 1,125Senior Cloud Architects, Engineering, & Technology Executives, with a minimum seniority of Director in 3 regionsacross 11 markets: North America (U.S., Canada), EMEA (Germany, Italy, France, UK, Israel), and APAC (India,Australia, Singapore, Japan), between December 5th and December 16th, 2025, using an email invitation and anonline survey. Results of any sample are subject to sampling variation. The magnitude of the variation is measurable and isaffected by the number of interviews and the level of the percentages expressing the results. For the interviewsconducted in this particular study, the chances are 95 in 100 that a survey result does not vary, plus or minus, bymore than 3.1 percentage points in the global sample, 6.9 percentage points in the United States, 19.6 percentagepoints in Israel, and 9.8 percentage points in each of the remaining markets from the result that would be obtainedif interviews had been conducted with all persons in the universe represented