Alibaba's latest artificial intelligence strategy signals a shift from competing primarily through individual models to building an integrated technology system capable of supporting increasingly demanding AI applications. The company plans to develop future models with between 5 trillion and 10 trillion parameters while simultaneously introducing a new AI chip and expanding its global data centre capacity. The significance of the announcement lies not only in the size of the proposed model, but in the infrastructure Alibaba is building around it.
The company announced the plans at its annual Apsara Conference in Hangzhou, where Chief Executive Eddie Wu outlined a strategy covering AI models, semiconductors, computing infrastructure and cloud services. Alibaba's next-generation Qwen 4 model is already in training, while later Qwen 4.5 and Qwen 5 models are projected to reach the 5 trillion to 10 trillion parameter range. That would represent a substantial increase from the 2.4 trillion parameters of the company's current Qwen 3.8 Max flagship model.
The announcement comes at a time when the economics of advanced AI are increasingly determined by access to computing power. Larger models require enormous quantities of processing capacity, memory, electricity and data centre infrastructure. Alibaba's decision to address these requirements simultaneously suggests that the company sees control over the entire technology stack as increasingly important to its ability to compete.
Bigger Models Require More Than More Parameters
The proposed increase in model size is significant, but parameter count alone does not determine an AI system's usefulness. Alibaba's Qwen 3.8 Max already uses a sparse mixture-of-experts architecture, allowing the model to contain 2.4 trillion parameters while activating only a much smaller portion during individual tasks. This approach demonstrates why future progress will depend not simply on making models larger, but on improving how efficiently their computing resources are used.
Alibaba says its future models are intended to handle more complex and longer-running tasks. That points towards a broader change in the role of AI systems, from answering individual prompts towards completing multi-stage assignments involving planning, reasoning, coding, research and interaction with other software.
The company has also highlighted progress in what it describes as recursive self-improvement. The concept involves AI systems identifying limitations, designing experiments and generating additional training material to improve their performance. If such techniques become practically reliable, they could influence how quickly future models improve without requiring every advance to come from conventional human-designed training processes.
However, the proposed 5 trillion to 10 trillion parameter range should be viewed as a development target rather than evidence that such a model already exists. Alibaba has not indicated that a model of that scale has been completed, and the technical and financial requirements for training one remain substantial.
The New Chip Addresses a Strategic Bottleneck
The introduction of the Zhenwu V900 AI chip is therefore closely connected to the model announcement. Alibaba is not simply planning a larger model while depending entirely on external semiconductor suppliers. Its T-Head semiconductor unit is developing its own computing hardware, giving the company greater control over an important part of the AI development chain.
Alibaba says the V900 provides three times the performance of its predecessor, the M890, which was introduced only months earlier. The company also says that clusters based on the new chip can scale to as many as 500,000 units for advanced model training and inference. Mass production and commercial availability are planned for the first quarter of 2027.
That rapid progression illustrates the pressure facing Chinese technology companies as they attempt to increase domestic computing capacity. Restrictions on the export of advanced computing technology to China have made access to some of the world's most sophisticated AI processors more difficult. Domestic chip development has consequently become both a commercial objective and a strategic technology priority.
Yet developing an AI accelerator is only one part of solving the problem. Competitive performance depends on manufacturing capability, advanced packaging, memory, networking, software optimisation and the ability to operate thousands of processors together efficiently. A new chip can therefore strengthen an AI ecosystem without automatically eliminating its dependence on the wider semiconductor supply chain.
Data Centres May Determine the Pace of Expansion
Alibaba's plan to increase its global data centre capacity beyond 20 gigawatts by 2032 shows the scale of infrastructure it believes will be required. The company has acknowledged that demand for AI computing is growing faster than its ability to supply it, with shortages affecting several parts of the data centre industry.
This is an important limitation because advanced AI models cannot be separated from physical infrastructure. Training and operating very large models require electricity, cooling systems, networking equipment, storage and specialised computing facilities. As models become larger and AI agents perform longer sequences of tasks, demand for computing can increase even when the number of individual users does not rise at the same rate.
Alibaba's position as both a cloud provider and an AI developer gives it a particular reason to invest heavily in this infrastructure. Computing capacity can support the company's own model development while also being sold to businesses through cloud services. This creates a potential commercial link between the enormous cost of AI development and the revenue generated by customers using AI computing.
The strategy also means that Alibaba is positioning cloud infrastructure as more than a hosting business. If advanced AI becomes increasingly dependent on specialised computing environments, ownership of those environments could become a source of competitive advantage for cloud providers.
China’s AI Competition Is Becoming More Integrated
Alibaba's approach reflects a broader pattern in China's technology industry, where companies are attempting to develop domestic alternatives across models, chips and computing infrastructure. The objective is not simply to produce a competitive chatbot or language model but to create an ecosystem that can continue developing despite restrictions on access to some foreign technologies.
Competition within China is also intensifying. Alibaba faces pressure from established technology companies as well as newer AI developers that have demonstrated strong performance with comparatively efficient models. The emergence of powerful domestic alternatives has made efficiency particularly important because access to the most advanced computing hardware remains constrained.
This makes Alibaba's emphasis on its own chips and cloud infrastructure significant. The company is effectively trying to reduce the number of external dependencies that could slow the development of future AI systems.
The strategy nevertheless carries considerable costs. Building large models, designing advanced processors and expanding data centres require sustained investment before commercial returns can be guaranteed. The company must therefore convert technological capacity into cloud revenue, enterprise applications and other businesses capable of supporting those investments over time.
The Real Test Will Be Efficiency and Commercial Use
Alibaba's announcement ultimately shifts attention away from the headline figure of 10 trillion parameters towards a more practical question: whether the company's expanding infrastructure can translate into useful and commercially sustainable AI systems.
A model several times larger than Qwen 3.8 Max could require substantially greater computing resources, but larger does not automatically mean better for every application. Improvements in architecture, training methods, data quality and inference efficiency can sometimes deliver significant gains without proportional increases in model size.
Alibaba therefore has to solve two problems simultaneously. It must develop models capable of handling increasingly complex tasks while also making the underlying computing infrastructure efficient enough to operate them at commercial scale.
The planned Zhenwu V900, the expansion of Alibaba Cloud and the future Qwen models form parts of that same strategy. Rather than treating AI models, chips and data centres as separate investments, Alibaba is attempting to connect them into one development cycle in which better hardware enables larger models, larger models increase cloud demand, and greater cloud demand helps justify further infrastructure investment.
That approach could become increasingly important as the AI industry moves beyond the initial race to produce increasingly capable models. The next phase is likely to depend as much on the ability to train, deploy and operate those systems economically as on achieving another increase in model size. Alibaba's latest strategy suggests that the company is preparing for that infrastructure-intensive phase of the AI industry.
(Source:www.nbcnews.com)
The company announced the plans at its annual Apsara Conference in Hangzhou, where Chief Executive Eddie Wu outlined a strategy covering AI models, semiconductors, computing infrastructure and cloud services. Alibaba's next-generation Qwen 4 model is already in training, while later Qwen 4.5 and Qwen 5 models are projected to reach the 5 trillion to 10 trillion parameter range. That would represent a substantial increase from the 2.4 trillion parameters of the company's current Qwen 3.8 Max flagship model.
The announcement comes at a time when the economics of advanced AI are increasingly determined by access to computing power. Larger models require enormous quantities of processing capacity, memory, electricity and data centre infrastructure. Alibaba's decision to address these requirements simultaneously suggests that the company sees control over the entire technology stack as increasingly important to its ability to compete.
Bigger Models Require More Than More Parameters
The proposed increase in model size is significant, but parameter count alone does not determine an AI system's usefulness. Alibaba's Qwen 3.8 Max already uses a sparse mixture-of-experts architecture, allowing the model to contain 2.4 trillion parameters while activating only a much smaller portion during individual tasks. This approach demonstrates why future progress will depend not simply on making models larger, but on improving how efficiently their computing resources are used.
Alibaba says its future models are intended to handle more complex and longer-running tasks. That points towards a broader change in the role of AI systems, from answering individual prompts towards completing multi-stage assignments involving planning, reasoning, coding, research and interaction with other software.
The company has also highlighted progress in what it describes as recursive self-improvement. The concept involves AI systems identifying limitations, designing experiments and generating additional training material to improve their performance. If such techniques become practically reliable, they could influence how quickly future models improve without requiring every advance to come from conventional human-designed training processes.
However, the proposed 5 trillion to 10 trillion parameter range should be viewed as a development target rather than evidence that such a model already exists. Alibaba has not indicated that a model of that scale has been completed, and the technical and financial requirements for training one remain substantial.
The New Chip Addresses a Strategic Bottleneck
The introduction of the Zhenwu V900 AI chip is therefore closely connected to the model announcement. Alibaba is not simply planning a larger model while depending entirely on external semiconductor suppliers. Its T-Head semiconductor unit is developing its own computing hardware, giving the company greater control over an important part of the AI development chain.
Alibaba says the V900 provides three times the performance of its predecessor, the M890, which was introduced only months earlier. The company also says that clusters based on the new chip can scale to as many as 500,000 units for advanced model training and inference. Mass production and commercial availability are planned for the first quarter of 2027.
That rapid progression illustrates the pressure facing Chinese technology companies as they attempt to increase domestic computing capacity. Restrictions on the export of advanced computing technology to China have made access to some of the world's most sophisticated AI processors more difficult. Domestic chip development has consequently become both a commercial objective and a strategic technology priority.
Yet developing an AI accelerator is only one part of solving the problem. Competitive performance depends on manufacturing capability, advanced packaging, memory, networking, software optimisation and the ability to operate thousands of processors together efficiently. A new chip can therefore strengthen an AI ecosystem without automatically eliminating its dependence on the wider semiconductor supply chain.
Data Centres May Determine the Pace of Expansion
Alibaba's plan to increase its global data centre capacity beyond 20 gigawatts by 2032 shows the scale of infrastructure it believes will be required. The company has acknowledged that demand for AI computing is growing faster than its ability to supply it, with shortages affecting several parts of the data centre industry.
This is an important limitation because advanced AI models cannot be separated from physical infrastructure. Training and operating very large models require electricity, cooling systems, networking equipment, storage and specialised computing facilities. As models become larger and AI agents perform longer sequences of tasks, demand for computing can increase even when the number of individual users does not rise at the same rate.
Alibaba's position as both a cloud provider and an AI developer gives it a particular reason to invest heavily in this infrastructure. Computing capacity can support the company's own model development while also being sold to businesses through cloud services. This creates a potential commercial link between the enormous cost of AI development and the revenue generated by customers using AI computing.
The strategy also means that Alibaba is positioning cloud infrastructure as more than a hosting business. If advanced AI becomes increasingly dependent on specialised computing environments, ownership of those environments could become a source of competitive advantage for cloud providers.
China’s AI Competition Is Becoming More Integrated
Alibaba's approach reflects a broader pattern in China's technology industry, where companies are attempting to develop domestic alternatives across models, chips and computing infrastructure. The objective is not simply to produce a competitive chatbot or language model but to create an ecosystem that can continue developing despite restrictions on access to some foreign technologies.
Competition within China is also intensifying. Alibaba faces pressure from established technology companies as well as newer AI developers that have demonstrated strong performance with comparatively efficient models. The emergence of powerful domestic alternatives has made efficiency particularly important because access to the most advanced computing hardware remains constrained.
This makes Alibaba's emphasis on its own chips and cloud infrastructure significant. The company is effectively trying to reduce the number of external dependencies that could slow the development of future AI systems.
The strategy nevertheless carries considerable costs. Building large models, designing advanced processors and expanding data centres require sustained investment before commercial returns can be guaranteed. The company must therefore convert technological capacity into cloud revenue, enterprise applications and other businesses capable of supporting those investments over time.
The Real Test Will Be Efficiency and Commercial Use
Alibaba's announcement ultimately shifts attention away from the headline figure of 10 trillion parameters towards a more practical question: whether the company's expanding infrastructure can translate into useful and commercially sustainable AI systems.
A model several times larger than Qwen 3.8 Max could require substantially greater computing resources, but larger does not automatically mean better for every application. Improvements in architecture, training methods, data quality and inference efficiency can sometimes deliver significant gains without proportional increases in model size.
Alibaba therefore has to solve two problems simultaneously. It must develop models capable of handling increasingly complex tasks while also making the underlying computing infrastructure efficient enough to operate them at commercial scale.
The planned Zhenwu V900, the expansion of Alibaba Cloud and the future Qwen models form parts of that same strategy. Rather than treating AI models, chips and data centres as separate investments, Alibaba is attempting to connect them into one development cycle in which better hardware enables larger models, larger models increase cloud demand, and greater cloud demand helps justify further infrastructure investment.
That approach could become increasingly important as the AI industry moves beyond the initial race to produce increasingly capable models. The next phase is likely to depend as much on the ability to train, deploy and operate those systems economically as on achieving another increase in model size. Alibaba's latest strategy suggests that the company is preparing for that infrastructure-intensive phase of the AI industry.
(Source:www.nbcnews.com)