Here is the English translation:
The important position of top cloud vendors like Microsoft (MSFT) and Amazon (AMZN) in the value chain continues under the Enterprise AI theme. Although some short-term financial data underperformed expectations, it does not affect their medium and long-term value.
-
In the software sector, companies like ServiceNow (NOW), Palantir (PLTR), Cloudflare (NET), and Elastic (ESTC) are worth investing in as they can help enterprises effectively deploy GenAI and have already started positive business transformation under the AI theme.
01. The penetration speed of GenAI on the enterprise side is faster than imagined
1. The speed of enterprise deployment of GenAI is accelerating
Although the cycle of training and releasing new generations of SOTA models is lengthening, large enterprises are not slowing down in adopting AI.
2023 was the year with the fastest penetration rate of AI on the enterprise side. In McKinsey's survey, the penetration rate of AI on the enterprise side increased from 55% to 72%, an increase of 17 percentage points. If AI is narrowed down to GenAI, the speed is even more astonishing, growing from 33% to 65% in the past year, doubling.
Enterprise budget allocation for AI is also increasing. According to Morgan Stanley's 2024Q2 US Tech Report, ### in Q2 2024, the growth rate of enterprise AI/ML related project budgets is 16.3%, compared to 13.7% in Q1. Among the top 10 enterprise project expenditures, only CRM Application (+2.3 percentage points) and Storage Hardware Data (+1 percentage point) did not see a decline in growth rate, and these two sectors are also key facilities for enterprise AI deployment.
2. 2025 will see massive adoption of GenAI on the enterprise side
Even though all surveyed enterprises have recognized the importance of GenAI, companies of different sizes are at different stages of AI deployment.
According to UBS's survey on enterprise AI spending, the majority of medium and large enterprises are mainly focused on researching use cases for proof of concept and small-scale test deployments of GenAI. Among them, 45% of large enterprises have started small-scale test deployments, and 40% have identified their use cases and started proof of concept. For medium-sized enterprises, these two stages account for 44% and 38% respectively.
Small enterprises, on the contrary, are concentrated at both ends. In the survey, 25% of small enterprises have already entered the stage of large-scale deployment, while another 25% are still in the research stage. The higher percentage of small enterprises able to quickly scale AI deployment may be related to factors such as decision-making flexibility due to smaller size, relatively uncomplicated workflows, and higher cost sensitivity. The 25% still in the research phase may be related to the enterprise's technical reserves and its own business development roadmap.
Overall, we can optimistically expect that ### by the end of 2024, more medium and large enterprises will expand the deployment scale of GenAI within their organizations, and even begin to apply it to broader business flows (In production at scale across units). 2025 will see massive adoption of GenAI on the enterprise side.
If we fail to correctly recognize the "progressive unlocking" characteristic of models, we will overestimate the short-term progress in model capabilities and the rate at which LLMs truly impact actual business. This is also reflected in the changing expectations of CIOs regarding the timeline for AI/LLMs to be actually applied to enterprise production. From Q3 2023 to Q2 2024, CIOs' estimated timelines for using AI have clearly slowed down.
In the Q4 2023 survey, the market was most optimistic about the application cycle of GenAI: 1/3 of respondents believed that GenAI could be used in actual business production (in production) after half a year, while 1/3 of enterprises indicated they had no plans on how to use GenAI.
By Q1 and Q2 2024, enterprises' judgments on the timeline have become cautiously optimistic. In the latest Q2 2024 survey, among the respondents, 26% expect to see AI/LLMs enter enterprise production processes after 2025, while 25% believe it will happen in H2 2024.
3. In the short to medium term, enterprise GenAI use cases are still concentrated in internal scenarios
Over the past 4 quarters, enterprise users' views on how to use AI/LLM internally have also been changing:
• Enterprises targeting internal productivity improvement as the goal of AI/LLM deployment increased from 15% to 23%, rising from 3rd to 1st place;
• Enterprises using AI/LLM to optimize labor costs (e.g., simplifying business processes in customer service, finance, and other sectors) increased from 10% to 18%;
• Although still among the top 3 directions for AI/LLM deployment, the expectation of improving customer satisfaction decreased from 19% to 15%, dropping from 1st to 3rd place.
This change is quite interesting, as enterprise expectations for GenAI have shifted from external, front-office business scenarios to internal cost reduction and efficiency improvement, indirectly reflecting which scenario tasks have been more effective in enterprise AI/LLM attempts over the past 9 months. Under this consensus, we expect that in the short to medium term, enterprises will continue to use AI more in internal scenarios.
Although the potential of Gen-AI is widely recognized, Bain's survey shows that currently only about 35% of companies can clearly describe how to create business value from Gen-AI. The complete transition from exploration stage to large-scale implementation may not be as rapid as we expect, but rather a gradual process over 3-5 years.
To quantify the potential value of GenAI, McKinsey chose to analyze different enterprise GenAI use case scenarios by predicting the impact amount of GenAI and the proportion of impact on functional spend.
Among all functional sectors, GenAI has the most significant impact on Sales & Marketing, Software Engineering, Corporate IT, Customer Operations, and Product R&D. These sectors account for about 75% of the total annual impact of GenAI on enterprises.
In addition to the obvious concentration of use cases, we also found that customer service, software development, enterprise IT