日本IT服务业NEC与富士通将驱动企业AI商业化维持买入评级

Japan IT services NEC and Fujitsu set to drive enterprise AI monetization, Staying Buy CITI'S TAKE Masahiro ShibanoAC+81-3-6776-4641masahiro.shibano@citi.com We believe NEC and Fujitsu share prices still reflect AI substitutionconcerns. However, we still regard both companies as essential for longer-term enterprise AI implementation. Recent cases from AI implementationfrontrunners suggest concerns about AI ROI are increasing. As frontier AImodel token prices continue rising, enterprises must implement AI whileconsidering the attendant economics. As Microsoft's CEO stated, enterpriseAI implementation requires an AI ecosystem, and we believe NEC andFujitsu play essential roles in building enterprise AI ecosystems. Moreover,growth in their respective hardware domains has been remarkable. We referhere in particular to NEC's defense, space, and submarine cables, andFujitsu's optical transmission, semiconductors, and quantum computing. Moriya Koketsu+81-3-6776-8304moriya.koketsu@citi.com AI adoption—In Japan, full-scale AI implementation remains a medium- to long-term theme. Japanese companies remain focused on modernizing core systems,withrelated demand continuing through CY28-CY29.Meanwhile,leadingcompanies, primarily in the US, have already advanced AI implementation withpioneering case studies emerging. The high AI costs US firms currently face areprompting ROI discussions. The AI utilization phase at leading companies is shiftingfrom maximization to optimization. RoI—Comments on AI RoI by Uber’s COO, Microsoft's suspension of certain AI tools,and Meta and Amazon's shift from maximizing to optimizing AI tool usage suggestcurrent AI utilization does not necessarily generate returns commensurate withinvestment. Importantly, similar cases have been seen even in the coding domainwhere returns were considered high. SIer positioning—Evidence to date thus suggests the need to use multiplepurpose-specific AI models. To generate RoI from AI investment, companies mustdeploy AI models with appropriate performance and usage fees based on businessvalue and requirements. This suggests enterprise AI deployment is set to becomeincreasingly complex, which we consider positive for NEC and Fujitsu, which provideAI deployment support. nEarnings forecasts—For NEC, we forecast FY3/27 non-GAAP OP of ¥454bn(YoY+14%). Despite the dropout of extraordinary public IT services demand, weexpect OP to beat guidance (¥420bn) due to strong ANS (defence, aerospace,submarine cables) profits. For Fujitsu, we forecast FY3/27 adjusted OP of¥415.9bn (YoY+7%) versus guidance of ¥425bn. We expect hardware solutionsand ubiquitous solutions to undershoot guidance due to higher memory costs. AI and RoI RoI by area Having begun with AI utilization based on chatbots and coding by engineers,corporate AI adoption is now shifting to the implementation of AI agents across awide range of business domains. As implementation is applied more broadly, a keydiscussion is whether "the economic value of the work justifies the AI cost." Forexample, the usage fees for an AI agent in the legal domain are highly likely to bejustified by the economic value of the work in that domain. The regulatory industry,including lawyers, is short-staffed, and the cost per person-month is high.Furthermore, many legal tasks are rule-based, making it relatively easy to confirmthe accuracy of AI processing. Therefore, the RoI of automating these tasks with AIis high. Call center operations are another area where RoI is relatively high. Becausethe work involved is chiefly manual-based responses and is standardized, theautomation effect can easily scale with the large number of cases, even if the costper person-month of the personnel being replaced is relatively low. Is AI RoI high in all areas? The first area is repetitive work handled by specialized personnel with high person-month costs. This includes some tasks such as legal affairs, regulatory compliance,financial analysis, and development support. In these areas, the person-monthcost of the personnel to be replaced or supplemented is high, so the economicvalue created by AI-driven efficiency is also large. In addition, if the work is rule-based and repetitive to a certain extent, it is easier for humans to re-verify theresults after AI processing, and the implementation risk is easier to manage. The second is work where a large volume of standardized tasks exists, even thoughthe person-month cost is relatively low. This includes routine processing in callcenters and back offices. In this case, although the economic value per case is notlarge, the automation effect of AI is easy to scale due to the large number of casesprocessed. On the other hand, there are also clear areas where AI RoI is low. Even if the work ishandled by specialized personnel with a high person-month cost, if tasks are non-repetitive and judgment verification is difficult, the cost of verifying AI outputbecomes high. Furthermore, for low-volume an