2030年SaaS与生成式AI：软件并非“消亡”，而是蓬勃发展

SaaS vs. GenAI in 2030: software isn't "dead", it thrives A thesis exists that vendor sold SaaS will wither away in a GenAI world, either due to directreplacement by Agents, budget displacement, or increased DIY. We think these worriescan’t be further from the truth: SaaS vendors may be even harder to replace. Not to mentionIT / customer spend could see massive tailwinds from the value added, causing SaaSvendors to gather MORE value, not less in a GenAI world. BUT! We also agree there will be alarge change in where the effort and value-add of vendors is derived. We don’t expect all ofthis value to go to existing vendors as we are also likely to get massive new winners. Peter Weed+1 917 344 8390peter.weed@bernsteinsg.com Luwei Yang+1 917 344 8342luwei.yang@bernsteinsg.com Armin Hadavi, CFA+1 917 344 8463armin.hadavi@bernsteinsg.com Where we most agree with the Agents replace SaaS bear narrative: End User Tools.Agents threaten to reduce the need or fully replace traditional End User Tool softwareused to speed up human creative tasks (e.g., written content, graphics, analytical queries,software code). In addition, some jobs may partially or fully wither away, vastly (or nearlytotally) reducing seat count....but Agents probably don’t replace “programmatic” LoBor infrastructure operations SaaS (they may enhance them, instead).Areas wheretraditional SaaS is a better fit than Agents are in cases where you want exact, predictableoutput, high performance at low cost, etc. Instead, Agents are naturally a value-add ontop. Agents also may scale the amount of teamwork going on within the LoB application,improve adherence and data entry quality to processes, etc. Agents also likely increase thescale of infrastructure operations software needed. A new opportunity: DIY Agents are hard, thus Agent SaaS offerings proliferate.Tobuild an AI-native product, one where Agents are part of the core, is HARD and requires anearly continuous level of tending! We are talking about getting the Agents to do what youwant, reliably, effectively, and cheaply. The unfortunate memory limitations of LLMs andtheir cost to operate also mean getting the Harness and Scaffolding right is the differencebetween an Agent working long term and not, and getting the ROI on Agent operating cost. The open questions include: will spending on Agents pressure SaaS spending?Longterm, we see the size of the pie growing due to Agents. But CIOs need to find those dollarsto afford AI somewhere. What we expect: spend being displaced to afford / justify AI spendis generally not the SaaS LoB or infrastructure operations vendors....and who gets paidfor the Agentic value (new entrant or incumbent)?We think existing vendors are likelyto capture some to most spend. There are two scenarios where a new entrant is likely todisproportionately succeed: when the buyer changes / new buyer emerges, or when a“Switzerland” makes more sense. DIY software is an unlikely threat to most SaaS categories.If you plan to DIY, you’dneed to invest in R&D and operations (e.g., hosting/scaling, cybersecurity) expertise on thattopic like you are the vendor yourself. This permanent, costly, and high-effort investmentgenerally only makes sense in the most strategic areas of a business.Not to mention,SaaS value is going up vs. DIY.Cycle times are compressing, capability expectations aregoing up, and adding Agent capabilities that actually work reliably is HARD. SaaS vendorsare more likely to have the critical human ingredients of Rigor and Judgment necessary forsuccessful use of AI Coding Copilots to write value-add code. Not to mention, downstreamDevOps/CloudOps is getting overwhelmed from increased coding velocity… SaaS vendoroperational sophistication may further outpace DIY. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS No change to Models, Price targets, or Recommendations. INFRASTRUCTURE, DEVELOPER, AND IT SAAS [IN ORDER OF GEN AI / AGENTIC TAILWIND VS. RISK… MORE UPSIDE AT THE TOP, MORE DOWNSIDE AT THEBOTTOM] Large potential upside, more limited downside: ServiceNow (NOW) LoB platform protected value, natural Agent-upsell upside, AND a short list of winners of the largest opportunity: Agent platform.ServiceNow is playing for one of the largest upsides — the new operating system for agents in large enterprises. With its current market-leading position in automating and managing business workflows end-to-end in large organizations,ServiceNow is very well positioned to be an Enterprise AI Agent platform. As we have discussed in our previous note,ServiceNow’s position as a “Switzerland” AI Agent platform remains durable and valuable against model providers. The CMDB,process rails, organizational context, and integrations with the broader ecosystem that ServiceNow has built over the years andmade them successful will continue to be a valuable moat. While agents cannot replace ServiceNow’s deterministic workflows,they naturally enhance the less programmatic parts of the workflow