Alibaba's latest results tell two very different stories about artificial intelligence. The first is a growth story. Revenue increased by approximately 9% year-on-year in the three months to June, while Alibaba's AI cloud and compute services revenue grew by an impressive 45%.
The second is a cost story. Quarterly net profit fell by approximately 75%, while capital expenditure increased by 75% to RMB67.68 billion, that’s almost $10 billion in just three months. Much of that investment is being directed towards the infrastructure required to support Alibaba's enormous AI ambitions.
Alibaba has already spent around half of the RMB380 billion (approximately $56 billion) it plans to invest in AI and cloud infrastructure between 2026 and 2029.
That's a staggering investment. It should make business leaders think carefully about the economics of their own AI strategies. Because there is a paradox emerging at the heart of enterprise AI: Businesses are investing in AI to reduce costs. But poorly planned AI programmes could end up doing exactly the opposite, turning their projects into an "AI money pit".
It would be easy to look at Alibaba's 75% fall in quarterly profit and conclude that its AI strategy is failing. That would miss the point. Alibaba is making a deliberate strategic investment. Demand for its cloud infrastructure is growing rapidly. AI-related products have delivered triple-digit revenue growth for 12 consecutive quarters, according to the company's latest results, while AI cloud and compute services revenue grew 45%.
Alibaba CEO Eddie Wu believes the company could break even on its AI-related capital expenditure within three years at current average gross margins. Alibaba is effectively investing today's profits in tomorrow's infrastructure. And it has good reasons for doing so. Alibaba isn't simply using AI. It wants to become one of the companies supplying the infrastructure, models and computing power upon which other organisations build their AI strategies. For Alibaba, AI infrastructure is becoming part of the product. For most enterprises, it isn't. And that's an important distinction.
Most CIOs aren't trying to build the next global AI platform. They're trying to use AI to improve their existing organisation. The business case normally contains some combination of:
Ultimately, the promise is usually more capability for less cost. But AI itself isn't free.
Enterprise adoption can require new infrastructure, cloud capacity, licences, specialist skills, integrations, consultants, governance, security controls and considerable organisational change. And that creates a financial risk which deserves more attention. A programme intended to reduce OPEX can start consuming OPEX. A programme intended to avoid infrastructure costs can start driving CAPEX. A programme designed to eliminate manual intervention can create another layer of specialist administration. And a programme intended to accelerate transformation can become another multi-year transformation programme. The danger isn't investment. The danger is investment without return.
Alibaba isn't an isolated example. Technology companies around the world are committing extraordinary amounts of capital to AI infrastructure. The largest hyperscalers are investing hundreds of billions of dollars in data centres, processors, networking and power infrastructure required to train and operate increasingly sophisticated AI systems. Reuters reported in August that investors are increasingly asking not simply who is spending the most on AI, but which businesses will ultimately generate sufficient returns from that investment.
That distinction matters. Alibaba, Microsoft, Amazon and Google can invest at extraordinary scale because they expect to sell the infrastructure and services being created. For the enterprise CIO, the equation should be much simpler:
What measurable business value will this AI investment produce?
And that creates a second question which should come before it:
What could prevent us from realising that value?
One answer is already sitting inside many organisations.
Their data.
AI is extraordinarily capable. But it still has to work with the information it is given. If your organisation has duplicate asset records, AI sees duplicate records. If your CMDB contains stale information, AI consumes stale information. If application dependencies are missing, AI doesn't automatically know they exist. If discovery coverage is incomplete, AI operates with incomplete visibility. If five different operational systems describe the same asset differently, the AI inherits that ambiguity. And if nobody inside the organisation knows which system should be trusted, AI doesn't magically resolve the governance problem. This matters because many of the most compelling enterprise AI use cases depend on operational data. Consider an AI agent being asked:
"Which infrastructure can we safely decommission?"
Or:
"Which applications are affected by this server failure?"
Or:
"Where can we reduce software licensing costs?"
Or:
"Which assets represent the greatest operational risk?"
Or:
"What caused this service outage?"
The intelligence of the model is only one part of the equation. The other is whether the underlying information accurately represents the organisation. AI can process bad data faster. It cannot turn bad data into operational truth.
TekWurx has seen what happens when organisations invest heavily in technology transformation before the underlying operational challenges have been resolved.
At a major international Stock Exchange, a strategic decision was made to migrate from BMC Discovery to ServiceNow Discovery. The project was important. The organisation needed to retire BMC Discovery and move to its strategic platform.
But after two years and four unsuccessful attempts with different partners, the migration was still failing. ServiceNow Discovery was identifying just 7% of the infrastructure previously identified by BMC Discovery. That meant most of the Stock Exchange's IT infrastructure remained hidden and unmanaged. The organisation couldn't confidently model its business applications and services. Integrations were failing. And delaying the retirement of the existing Discovery platform risked creating additional licensing costs. Think about the economics of those partners and implementation attempts over two years, the internal resources, consultancy costs. And technology costs. The level of management attention and the delayed benefits all for only 7% visibility.
This is what failed transformation looks like financially. The technology investment doesn't simply fail to generate its intended return. It starts creating additional cost.
TekWurx was ultimately asked to recover the programme. Within 152 days, the ServiceNow Discovery implementation was successfully completed. Discovery coverage increased from 7% to 98%. AWS and Azure assets were integrated into the CMDB. Business applications were linked directly to supporting infrastructure through tag-based mapping. Database visibility was established.
The enriched CMDB became a single source of truth used by IT owners, ITSM processes and asset lifecycle tools. The business impact went considerably further than fixing Discovery. Executives gained a more accurate view of infrastructure and applications. Manual effort and blind spots in asset management were reduced. The organisation established the clean, connected operational data foundation required for more advanced ITOM capabilities. The Stock Exchange case study describes that foundation explicitly as "AI ready".
That experience contains an important warning for today's AI programmes: Starting quickly isn't the same as reaching value quickly. Four attempts at the wrong approach were considerably more expensive than getting the foundation right.
The scale may be different, but the principle is identical. Imagine an organisation investing heavily in an enterprise AI programme. The licences have been purchased, then the organisation discovers that the operational information feeding those systems can't be trusted. Suddenly the AI programme has another dependency:
Fix the data.
The time and cost implications determine that the organisation hasn't generated a pound of the efficiency originally used to justify the investment. This is how an AI programme designed to reduce cost can end up increasing it.
There is another danger. Organisations may be tempted to proceed anyway. AI systems can generate answers even when the information underneath them is incomplete. That's precisely why data quality matters. A dashboard with missing information may visibly look incomplete. An AI assistant can produce a perfectly fluent answer based on the same incomplete information. That creates the potential for false confidence at machine speed.
As organisations move from AI assistants towards agentic AI capable of taking actions across enterprise systems, the consequences become more significant. If AI is going to recommend changes, automate remediation, rationalise applications, optimise software licences or make infrastructure decisions, businesses need confidence in the operational context informing those actions. Automation magnifies the quality of the foundation underneath it. Trusted data enables better automation. Untrusted data automates uncertainty.
The objective of uControl isn't to replace an organisation's existing operational technology estate. It's to make the information across that estate more trustworthy and usable. Enterprise operational data commonly exists across Discovery tools, ITSM platforms, CMDBs, asset repositories and other systems. Over time, that information becomes fragmented, duplicated and inconsistent. uControl is designed to address that operational data problem.
It ingests and reconciles information across different sources, helping organisations establish a trusted view of their technology estate before that information is consumed by automation and AI.
uControl Insights extends that visibility further by discovering IT, OT, IoT and cloud environments and building detailed asset intelligence across the estate. It can identify servers, endpoints, cloud resources, network devices and other technology assets, while adding hardware, software, lifecycle and relationship context.
uControl Insights already uses AI within that trusted operational context, from natural-language interrogation of inventory and risk scoring to scan diagnosis, anomaly detection and estate insights.
The principle is simple: Establish operational truth first. Apply intelligence second.
The most important question for CIOs in 2026 may therefore not be:
"Which AI platform should we buy?"
It may be:
"Is our organisation ready for AI to start making decisions?"
Alibaba can spend almost $10 billion in a quarter building AI infrastructure because it is making a calculated strategic bet on becoming one of the companies powering the AI economy. Most enterprises are making a different bet. They're investing in AI because they expect it to make their organisations more productive, efficient and competitive. That means return on investment matters enormously. The most expensive AI project may not be the one with the biggest initial investment. It may be the one you have to build twice.
TekWurx can help you understand whether the operational data across your Discovery, CMDB, ITSM and asset landscape is ready to support AI and automation.
Book a uControl demonstration and see how trusted operational data can help you build your AI strategy on facts rather than assumptions.