What technological advancements truly marked 2024, and which were just media noise? Between the tightening regulations around artificial intelligence, the arrival of dedicated chips in our pockets, and the structuring of an embedded AI market, the year reshuffled the cards for manufacturers, developers, and users. Here’s what the data allows us to measure.
Embedded AI vs. Cloud AI: Where the Technological Shift Happens in 2024
The most structuring distinction of 2024 concerns not a product, but an architecture. Two models compete to execute artificial intelligence tasks: remote processing (cloud) and local processing (edge AI). The table below summarizes the concrete differences between these approaches.
You may also like : The latest in women's entrepreneurship: initiatives, tips, and events to follow
| Criterion | Cloud AI | Embedded AI (edge AI) |
|---|---|---|
| Latency | Variable, depends on the connection | Almost instantaneous, processing on the device |
| Data Privacy | Data transmitted to remote servers | Data remains on the device |
| Offline Autonomy | None without connection | Functional without network |
| Available Computing Power | Very high (data centers) | Limited by the local chip (NPU/TPU) |
| Operating Cost | Usage-based billing (API, subscription) | Included in the hardware price |
In 2024, several manufacturers of smartphones, PCs, and connected objects integrated dedicated NPU and TPU capable of locally executing generative AI models. Content summarization, translation, photo editing, voice transcription: these tasks are gradually migrating from the cloud to the device.
This shift modifies the value chain. Software publishers who charged for access to cloud APIs are now facing embedded competition, free once the hardware is purchased. To keep track of these developments over time, the Neo News site for tech regularly covers announcements from manufacturers and chip benchmarks.
See also : The fascinating and complex missions of the locksmith in the 3rd arrondissement of Lyon
European AI Regulation: Constraints Changing Tech Products
The high-tech news of 2024 is not limited to product launches. The European AI Regulation (AI Act) has entered its final adoption phase, with a timeline for implementation based on risk levels.
AI systems classified as “unacceptable risk” are subject to a ban. Those deemed high risk face enhanced documentation and human oversight requirements. For generative AI intended for the general public, transparency requirements apply: indicating that content has been produced or modified by AI becomes mandatory.
Regulatory Convergence Outside Europe
The framework is not limited to the European Union. In the United States, the FTC has published guidelines on misleading advertising related to AI. In the UK, the ICO has strengthened its powers regarding the use of AI in financial services. Canada and Australia have initiated their own regulatory frameworks.
This global trend towards legal regulation of generative AI has direct consequences on marketed products. AI assistants integrated into operating systems, foundation model APIs, and AI features in applications must now incorporate compliance mechanisms from the design stage.
- Smartphone manufacturers are adding transparency notices when AI modifies a photo or generates text, to anticipate the requirements of the AI Act.
- Platforms offering generative AI APIs are now documenting the training datasets used, under regulatory pressure.
- Companies deploying public chatbots are integrating systems to identify generated content, sometimes referred to as “digital watermarking.”
Hardware Innovation 2024: What NPU Chips Change in Practice
The arrival of NPUs (Neural Processing Units) in consumer devices represents a hardware turning point. These chips, distinct from CPUs and GPUs, are optimized for neural network operations: matrix multiplication, rapid inference, low energy consumption.
On a smartphone equipped with a recent NPU, voice transcription works without an internet connection. AI photo editing (object removal, resolution enhancement) executes in seconds instead of routing through a remote server. On a laptop, automatic document summarization or real-time translation during a video conference become native functions.
Current Limitations of Local AI
Models executable locally remain smaller than those hosted in the cloud. A language model running on a smartphone NPU does not compete with the capabilities of a full foundation model accessible via API. Embedded AI excels at targeted tasks, not versatility.
Fragmentation is also an issue. Each manufacturer develops its own NPU with its proprietary SDK. Application developers must adapt their code to each architecture, which hinders adoption. However, this competition among manufacturers accelerates performance gains from one generation of chips to the next.
Sustainable Technologies and Green Cloud: An Innovation Axis Still Poorly Measured
Green cloud computing gained visibility in 2024, driven by the climate commitments of major tech companies. The idea: reduce the carbon footprint of data centers that power all digital services, including AI.
Explored avenues include liquid cooling of servers, renewable energy sourcing, and software optimization to reduce the number of calculations needed for the same task. Some cloud providers now display a carbon footprint indicator per request, aimed at client companies.
The paradox remains: the rise of generative AI, which is computation-intensive, increases the consumption of data centers at a time when the goal is to reduce it. Energy efficiency is improving, but the demand for computation is growing faster.
Thus, 2024 will have been a year of dual movement: technology is migrating to local devices with embedded AI, while regulation catches up with the most massive uses of the cloud. The products that will emerge in 2025 will bear the mark of these two simultaneous pressures, both material and legal.