Start-Ups Shift AI Tasks to Smartphones

Artificial intelligence is increasingly moving out of distant data centers and onto the devices people carry every day. A new wave of funding and product announcements shows that start-ups and major technology companies are racing to run more capable AI models directly on smartphones, laptops and other personal devices. The goal is not to replace the cloud entirely, but to shift many everyday AI tasks closer to the user, where they can be faster, cheaper and more private.

Funding signals investor appetite

Beijing-based ModelBest has closed a C+ Series funding round, bringing its total raised this year to 5 billion yuan, equivalent to about $740 million or £550 million. Investor Shanghai Mejoy Capital disclosed the completion in a social media post and said the round valued ModelBest at 20 billion yuan. The investor did not disclose the size of the latest tranche. The deal underscores how quickly capital is flowing toward companies that promise to run capable AI models directly on smartphones, laptops and other edge devices rather than relying entirely on remote cloud servers.

ModelBest is one of the more visible specialists in on-device AI. Its work reflects a broader strategic shift across the industry: companies continue to build enormous models with trillions of parameters, but they are also investing in smaller, efficient models and software techniques that can execute increasingly sophisticated tasks locally. The appeal is straightforward. Local processing can reduce latency, lower bandwidth and cloud costs, and keep sensitive data on the device. For consumers, that could mean faster responses, better offline functionality and stronger privacy protections. For companies, it could mean cheaper AI services and more control over how data moves.

Why edge AI is gaining ground

Today, some AI features already run on phones and laptops. Voice assistants can recognize wake words, cameras can apply computational photography, and keyboards can suggest text without sending every keystroke to a server. But more compute-intensive tasks often get routed through remote facilities. That adds latency, increases costs and raises privacy questions. Edge AI seeks to rebalance that equation by handling more work on the device itself.

A smartphone cannot match a data center for raw computing power, but advances in model compression, quantization, chip design and selective activation are narrowing the gap for many everyday tasks. Instead of sending every request to the cloud, a device can manage voice commands, image editing, summarization, translation and personal assistance locally, falling back to remote systems only when needed. This hybrid approach is becoming the default architecture for companies that want to combine the responsiveness of local processing with the knowledge and scale of cloud models.

In China, edge AI has an additional strategic dimension. The approach is seen as a way to take advantage of locally produced AI chips that are better suited to training and running lighter-weight models than the most advanced frontier products. That matters as domestic technology companies seek to build competitive AI ecosystems around available hardware and software stacks. ModelBest said in May that its billion-parameter on-device MiniCPM5-1B model was trained entirely on Huawei’s Ascend chips. It achieved top benchmark scores on an index for open-weight models under 2 billion parameters, according to the company. The claim highlights how smaller models, trained and optimized for local hardware, can deliver useful performance without depending on the largest cloud clusters.

New products target local intelligence

Shanghai-based StepFun, backed by Tencent Holdings, last week announced Step Edge, a family of on-device AI models. The company said the models offer low latency, local data processing and the ability to handle simple tasks without a network connection. The pitch is practical: users should not have to wait for a round trip to a server for basic requests, and they should be able to use AI features in places where connectivity is weak or unavailable.

A day later, StepFun announced the StepX Neo, a smartphone designed around autonomous AI agents built to interpret complex user requests and carry out tasks independently. The company did not provide a launch date or pricing. The StepX Neo concept points to a larger ambition. Device makers are not merely adding AI features to existing phones; they are exploring how handsets might be reorganized around AI agents that can plan, act and coordinate across apps. That vision depends on a combination of local models for speed and privacy and cloud models for heavy reasoning. The smartphone becomes a hub that decides where each task should run, rather than a thin terminal that sends everything to a remote server.

Apple's device-ready model strategy

Apple is pursuing a similar hybrid approach. It relies on third-party providers such as Google and OpenAI for high-end tasks while also working on on-device models. Last month, the company announced the 3 billion parameter AFM 3 Core and a more advanced 20 billion parameter model that activates between 1 and 4 billion parameters based on user tasks. Both are designed to run on devices.

The architecture illustrates a common industry tactic: keep a larger model in reserve but activate only the subset needed for a given request, reducing compute and memory demands. Apple is also evaluating technology from Silicon Valley firm PrismML, which claimed last week to have successfully run Alibaba’s 27 billion-parameter Qwen 3.6 model locally on an iPhone, according to reports. Alibaba and Baidu are providing technology for Apple’s AI products in China, the companies said last week.

The Apple example shows why the cloud is not disappearing. High-end tasks still demand substantial compute, and providers with large data-center footprints remain important. But the balance is changing. As on-device models improve, more functions can be handled locally, and the cloud is reserved for workloads that genuinely need it. That split could shape product design, app development and privacy policies for years.

Key facts at a glance

• ModelBest closed a C+ round and has raised 5 billion yuan, about $740 million or £550 million, so far this year.
• Investor Shanghai Mejoy Capital said the round valued ModelBest at 20 billion yuan, without disclosing the latest round size.
• ModelBest's MiniCPM5-1B model was trained entirely on Huawei Ascend chips and posted top scores on an open-weight benchmark for models under 2 billion parameters.
• StepFun announced Step Edge on-device models with low latency, local processing and offline capabilities for simple tasks.
• StepFun also announced the StepX Neo smartphone built around autonomous AI agents, without giving a launch date or price.
• Apple announced AFM 3 Core at 3 billion parameters and a 20 billion parameter model that activates 1 to 4 billion parameters depending on the task.
• Apple is evaluating PrismML technology that reportedly ran Alibaba's 27 billion-parameter Qwen 3.6 model locally on an iPhone.
• Alibaba and Baidu are providing technology for Apple's AI products in China.

Hardware, efficiency and the new competitive edge

The shift toward on-device AI is reshaping competition across the technology stack. Chipmakers are optimizing processors for matrix multiplication and other AI workloads while reducing power consumption. Model developers are experimenting with sparse activation, quantization, distillation and small-model architectures. Device makers are trying to balance memory, battery life and thermal limits. Software platforms must decide which tasks run locally, which run in the cloud and how to switch between them without disrupting the user experience.

These engineering choices have commercial consequences. A company that can run a useful model on a phone without draining the battery or heating the handset gains an advantage in responsiveness and privacy. A cloud provider that handles only the hardest requests can reduce infrastructure costs. A chip supplier that offers efficient inference for local models can win design slots in phones, PCs and wearables. Investors are watching these pieces fit together, which helps explain why a funding round for an on-device AI specialist attracts attention even in a crowded market.

There are also limits. Local models are typically smaller and less capable than the largest cloud systems. They may struggle with complex reasoning, long-context tasks or highly specialized knowledge. Updating them can be more complicated than updating a centralized service. Security remains a concern because models and data reside on devices that can be lost or compromised. And the split between local and remote processing creates new questions about consistency, governance and user expectations. Companies will need to be transparent about when data leaves the device and how AI decisions are made.

A market moving toward hybrid AI

The funding and product announcements point in one direction: the future of AI is likely to be hybrid rather than purely local or purely cloud-based. Smartphones and laptops will handle many routine and privacy-sensitive tasks on-device, while data centers will support demanding workloads, large-scale training and services that require up-to-date knowledge. This division could make AI feel more immediate and personal while keeping the heaviest computation where it is most efficient.

For start-ups, the opportunity lies in making small models genuinely useful. For established device makers, the challenge is integrating local AI without compromising battery life, performance or security. For chip suppliers, the prize is a larger role in the next generation of personal computing. The C+ round for ModelBest, the Step Edge launch, the StepX Neo concept and Apple’s device-ready models all show that the race is no longer only about who can build the biggest model. It is also about who can make powerful AI work quietly, quickly and privately on the devices people already own.

Source: Silicon UK News