In early June, Bain & Company released a report with a blunt title: "Your AI budget is increasing, but returns are not. Why." After reading the survey data from 951 large enterprises, I think the report's real value lies not in the clickbait conclusion that "AI is ineffective"—but in its clear explanation of a specific funding issue.
The issue is this: 44% of large enterprises are using the "expected savings" from the previous round of AI projects to support the budget for the next round of AI projects. And for 40% of the enterprises from the previous round, the actual savings amount was less than 10%.
Using money not yet received to buy the next batch of AI—this cycle will be clearly visible to the finance team on the 2027 financial statements.
Bain data: People with savings below 10% account for the largest share
In April 2026, Bain's Automation and AI Pathfinder Survey covered 951 companies worldwide with annual revenue exceeding $100 million, spanning nine industries including retail, financial services, manufacturing, and healthcare. Key data are as follows:
40% of AI projects achieve actual cost savings of less than 10% for enterprises. However, these enterprises' budgeted expected savings are 10% to 20%.
Saved less than 10%
40%
Enterprise
Save over 30%
4%
Enterprise
The middle 37% of progress is stuck between 10% and 20%—just enough to reach the number stated in the original proposal, but only barely meeting the target.
Bain’s diagnosis is clear: the bottleneck is not model capability but the data pipeline. No matter how fast the model runs, if it cannot read the master data in your ERP, or the customer information in your CRM is inconsistent, even the most powerful model can only make decisions based on incomplete information. MIT’s conclusion last year was the same—95% of generative AI pilot projects are stuck on the “pipeline” rather than the “engine.”
Bain's advice is very pragmatic: before the savings from the previous round of AI projects are actually realized, do not use them as a funding source for the next round of AI budget. This sounds like saying "don't borrow money to speculate in stocks"—logically correct, but in an environment where AI budgets grow by more than 20% per year in 2026, there are probably not many CFOs who can actually stop.
On the same day (almost simultaneously), Huawei Cloud and Deep Blue Technology presented two solutions
In the same week, as Bain's report sparked discussion, two domestic manufacturers separately released product roadmaps for enterprise AI deployment. I compared them and found that although their paths to solving the problem differ, they point to the same thing: reducing the cost of the gap between enterprise AI moving from “pilot” to “scale.”
Huawei Cloud: Don't compete on total token count; compete on the efficiency of running agents
On June 8, Zhou Yuefeng, CEO of Huawei Cloud, proposed a new concept called “Agentic Infra” at the INSPIRE Innovators Conference. He made a very interesting remark: “We don’t really care about the total number of tokens. Given that domestic computing power is indeed limited, we also don’t really care about total revenue.”
What he cares about is “whether the Tokens produced by domestic computing power can bring improvements in health and productivity, not just emotional value.” In other words: don’t just show me how many words the AI conversation generated—show me whether the defect rate on the production line has fallen.
The four products are released as a suite: AICS Lingqu Intelligent Computing Cluster (intelligent computing), AMS Memory Storage (memory), CCE Volcano Engine Next (scheduling), and AgentSphere (runtime environment). The enterprise entry point is named "Smart Orchard," designed for agents rather than humans.
"Industry AI Dream Factory" launches four major zones in its first phase—smart healthcare, embodied intelligence, smart manufacturing, and scientific computing. Healthcare was chosen as the pioneer field because among the 38,000 hospitals across China, only 5,000 have pathology departments. The Rui Pathology large model, developed in collaboration with Ruijin Hospital, has already been shared with more than 20 hospitals. The path is clear and explicit: start from the industry's deepest pain points, rather than from general capabilities.
DeepBlue Technology: Enabling One Person Plus a Group of AI Agents to Run the Entire Process
On June 4, DeepBlue Technology and China Mobile signed an agreement—China Mobile Cloud Phone will be equipped with the DeepAgent intelligent agent system. The core selling point of this system is “OPC lightweight single-person digital transformation”; put simply: one person coordinates, a group of AI agents executes, without changing systems, writing code, or adding staff.
From a technical perspective, there are several things worth carefully examining:
Operational scenario data: energy contract review shortened from 8 hours to 15 minutes (80% efficiency improvement), industry report generation efficiency increased by 200%, and overall operating costs can be reduced by up to 95%.
Interestingly, its pricing model is that the out-of-the-box SaaS version targets small and medium-sized enterprises, while the proprietary cloud and on-premises private deployment target governments, enterprises, and financial institutions. This is essentially the same logic as Huawei Cloud's “hybrid cloud + public cloud dual-track”: large customers need compliance, while small customers prioritize speed.
Let's take a look at these three clues together
The Bain report, Huawei Cloud INSPIRE, and DeepBlue DeepAgent—these events occurred in the same week, but it was no coincidence. Piecing them together reveals three real signals of enterprise AI implementation in June 2026:
Signal 1: ROI anxiety is spreading from the CIO level to the CFO level.Bain's data largely confirms one judgment: the AI investment returns of most enterprises have not reached the figures listed in their budgets. And the 2027 financial statements will bring this issue to the surface. The enterprises able to survive that round of scrutiny are likely to be those that have already begun, in accounting terms, to distinguish between "actual savings" and "expected savings."
Signal 2: The competitive focus of domestic vendors has shifted from "whose model is larger" to "who can make AI work through specific business operations." Huawei Cloud does not compete on total Token volume; instead, it focuses on pathology diagnosis and building an embodied intelligence development platform. DeepBlue is not showing off how strong its multimodal capabilities are; rather, it is telling you that contract review has been shortened from 8 hours to 15 minutes. This shift is healthy—after hearing about model parameters for over a year, enterprise customers now want to hear "how much manpower can I save?"
Signal Three: The barrier for SMEs to adopt AI is lowering, but the cost of choosing is rising. Lightweight solutions like DeepAgent that “require no system replacement and no coding” are aimed precisely at SMEs. Huawei Cloud’s Smart Orchard is doing the same thing—it is designed as an Agent service rather than a human service. But the problem is that there are now too many options: Huawei, DeepBlue, Alibaba Wuying, ByteDance’s Agent plan, Tencent WorkBuddy—what SME CIOs face is not “no available solutions,” but “not knowing which solution will still be usable in six months.”
Practical Operational Advice for Enterprise Digital Leaders
First, while there is still time, please separate "actual ROI" from "expected ROI" in your company's AI project statistics. The most dangerous figure in Bain's data is not the 40% of projects that missed targets, but the 44% of enterprises that use expected values to support their next round of investment. If your finance department does the same, please make a clear distinction now.
Secondly, before choosing an AI agent solution, conduct a data accessibility audit first. The core conclusion of Bain's diagnosis—that the bottleneck lies in the data pipeline rather than the model—has been corroborated by both Huawei Cloud and Shenlan's product positioning. Regardless of whom you choose, first ask whether the existing ERP/CRM/OA systems can allow AI to read the required data. If not, changing the model will be of no use.
Third, SMEs can consider a "lightweight + small-scenario entry" path. The interesting thing about DeepBlue's OPC model is that it does not require you to change your existing systems. If you are a manufacturing enterprise with annual revenue ranging from tens of millions to hundreds of millions, starting with single-point scenarios such as contract review, invoice processing, and report generation can produce results within three months, which is more prudent than directly adopting an "AI middle platform."
