Qualcomm CEO Says Smartphones Could Run 100-Billion-Parameter AI Models by 2028

Smartphones could become significantly more powerful artificial intelligence platforms over the next two years. Qualcomm CEO Cristiano Amon believes future smartphones may be capable of running AI models with 100 billion parameters directly on the device by 2028, potentially transforming how people interact with mobile technology.

Speaking about the future of mobile AI, Amon revealed that some leading artificial intelligence companies are discussing smartphones capable of running these large models continuously. He did not disclose the names of the companies involved.

The development could mark an important shift from today’s AI-powered smartphones, which often rely on smaller on-device models or cloud-based services to handle more demanding tasks.

However, major technical challenges remain, particularly in memory capacity, processing efficiency, battery consumption, and heat management.

Qualcomm CEO Reveals Ambitious Plans for On-Device AI

According to Amon, some AI companies want smartphones capable of running models containing at least 100 billion parameters by 2028.

Parameters are the numerical values an AI model learns during training. Although parameter count alone does not determine a model’s intelligence, larger models can offer greater capabilities when supported by suitable architectures, training, and computing resources.

Running a model directly on a smartphone is different from accessing an AI service over the internet. With on-device processing, the phone performs the AI computation locally instead of sending every request to a remote data centre.

This could allow future smartphones to offer more responsive AI assistants, improved privacy for certain tasks, and useful features even when an internet connection is unavailable.

Amon’s comments indicate an industry ambition rather than a confirmed product launch. Qualcomm has not publicly identified the AI companies involved or provided a complete technical roadmap showing how smartphones will meet the target.

Why 100-Billion-Parameter AI Models Are Difficult to Run on Phones

Today’s smartphones face substantial hardware limitations when running very large AI models.

One of the biggest obstacles is RAM, which temporarily stores the model’s parameters and the information needed to process requests.

For a conventional 100-billion-parameter model, memory requirements can quickly exceed what most smartphones currently offer.

Estimated Memory Requirements

These are simplified estimates based on the number of parameters and their storage precision. Actual requirements vary with the model architecture, quantization method, and additional storage overhead.

The figures also exclude memory needed for the operating system, applications, and the model’s working context.

For comparison, many current smartphones have substantially less RAM than these estimates suggest. Even a phone with 16 GB of RAM would struggle to accommodate a conventional 100-billion-parameter model in its entirety using these configurations.

Manufacturers would therefore need to combine hardware improvements with more efficient ways of storing and executing AI models.

Memory Costs Could Affect Future Smartphone Prices

Increasing RAM capacity is not simply a technical challenge. It can also increase manufacturing costs.

Memory prices and supply conditions influence the cost of producing smartphones, laptops, and other electronics. Equipping phones with substantially more memory could become expensive if manufacturers face tight supply or rising component prices.

This creates a difficult balance for smartphone companies: consumers want more capable AI features, but manufacturers must deliver them without making devices prohibitively expensive.

Qualcomm’s vision will therefore depend not only on faster processors but also on improvements in memory technology, model efficiency, and the overall cost of mobile hardware.

How Smartphone Makers Could Overcome the Limitations

Several technologies could help make large AI models more practical on smartphones.

1. AI Model Quantization

Quantization reduces the number of bits used to represent a model’s parameters, lowering memory requirements.

For example, storing parameters at four-bit precision requires considerably less memory than storing them at 16-bit precision.

More aggressive quantization could make larger models fit into limited memory. However, reducing precision too much can affect accuracy and reasoning performance, so developers must balance model size against quality.

2. Sparse Mixture-of-Experts Models

Another possible approach involves sparse Mixture-of-Experts, commonly known as MoE.

These models contain specialised components, or experts, but activate only a subset of them for each token. This can reduce the amount of computation required for a particular request compared with activating every component.

With suitable designs, some model components could remain in storage until needed, potentially reducing the amount of RAM required at any one time.

However, moving model components between storage and memory can introduce delays, making implementation and performance optimisation important.

3. Faster Memory and Improved Chip Design

More efficient processors, memory systems, and chip packaging could help future phones handle demanding AI workloads.

Manufacturers will need to improve how processors access model data while controlling power consumption and heat.

These advances could make sustained AI processing more practical, although they would not automatically eliminate the memory requirements of a large model.

4. Smaller Models With Advanced Training

Smartphone makers may not need to run every model at its full size.

Techniques such as distillation can transfer capabilities from a larger model into a smaller one. A well-designed smaller model may perform sufficiently well for many mobile tasks while consuming fewer resources.

For everyday activities, such as summarising messages, translating speech, or organising information, a smaller specialised model could be more practical than a full 100-billion-parameter system.

Heat and Battery Life Remain Major Challenges

Memory is only one part of the problem. Running a large AI model continuously could place substantial demands on a smartphone’s processor and battery.

Unlike desktop computers, smartphones have limited space for cooling. Prolonged processing can increase internal temperatures, potentially causing the device to reduce performance to protect its components.

Continuous AI activity could also shorten battery life, particularly if the model repeatedly processes audio, images, notifications, or other incoming information.

Manufacturers would need to improve energy efficiency and thermal management before large AI models could operate continuously without significantly affecting everyday smartphone use.

A successful solution must balance intelligence, responsiveness, battery life, and device temperature.

Could AI Companies Start Making Their Own Smartphones?

Amon’s comments also raise questions about the future relationship between AI companies and traditional smartphone manufacturers.

He indicated that some leading AI companies are exploring phones designed around more advanced on-device AI capabilities. However, he did not identify the companies or confirm which products they are developing.

If AI companies enter the smartphone market, they could compete by offering devices built around AI assistants and automated tasks rather than relying primarily on conventional apps.

Such devices might be designed to understand user requests, coordinate activities across applications, and provide more personalised assistance.

However, developing a successful smartphone requires more than an advanced AI model. Companies also need reliable hardware, an operating system, security protections, developer support, and a strong distribution and customer-service network.

For now, the possibility of AI companies launching their own smartphones remains an emerging industry direction rather than a confirmed market shift.

What Could This Mean for Smartphone Users?

If the industry overcomes the technical challenges, more powerful on-device AI could change how people use their phones.

Potential benefits include:

• More capable AI assistants: Phones could handle increasingly complex requests directly on the device.
• Improved privacy for some tasks: Processing sensitive information locally could reduce the need to transmit it to external servers.
• Faster responses: On-device processing could eliminate some delays associated with sending requests to the cloud.
• Better offline functionality: Certain AI features could remain available without an internet connection.
• More personalised experiences: Local AI systems could help users manage information and automate routine tasks.

These benefits will depend on the capabilities of the final models, how smartphone makers implement them, and which features they make available to users.

On-device AI will not necessarily eliminate cloud computing. Larger tasks may continue to require remote data centres, while smartphones handle routine activities locally.

Will Smartphones Really Run 100-Billion-Parameter Models by 2028?

Qualcomm’s target is ambitious, and several obstacles remain unresolved.

The company has not yet provided a detailed public roadmap explaining how future phones will combine sufficient memory, efficient processing, manageable temperatures, and acceptable battery life to run 100-billion-parameter models continuously.

Progress in quantization, sparse model architectures, storage technology, and specialised mobile processors could help close the gap.

It is also possible that future smartphones will use a combination of smaller local models and cloud-based AI instead of running every large model entirely on the device.

The key distinction is between the ability to execute a large model under certain conditions and the ability to run it continuously while maintaining normal smartphone performance.

Conclusion

Qualcomm CEO Cristiano Amon’s prediction points towards a future in which smartphones play a much larger role in artificial intelligence.

If manufacturers can overcome memory, cost, battery, and thermal limitations, future devices could support more advanced AI features with less dependence on cloud services.

However, 2028 remains a target rather than a guarantee. The eventual outcome will depend on advances in chip design, model architecture, memory technology, and energy efficiency.

For consumers, the most important change may not simply be larger AI models. It could be smartphones that understand requests more effectively, respond more quickly, and handle more everyday tasks directly on the device.

Editorial note: The 2028 timeline reflects Qualcomm CEO Cristiano Amon’s stated expectations. It should not be interpreted as confirmation that specific smartphone models or 100-billion-parameter on-device AI products will launch by that date.