The technological trends of 2024 are not just a list of buzzwords. Three structural changes are reshaping software and hardware architectures: the convergence of AI regulations on a global scale, the shift of AI computation to the endpoint, and the reconfiguration of trust chains around post-quantum cryptography. Here we detail the technical points that most annual overviews gloss over.
AI Regulations in 2024: Three Legal Frameworks, Three Implementation Logics
The European AI Act, definitively adopted in March 2024, imposes a phased implementation with obligations starting in 2025 for high-risk systems and the most powerful foundation models. For SaaS publishers and cloud providers, this means integrating “by design” compliance functions into the deployment pipeline, not just ticking a regulatory box afterward.
Read also : The Latest Trends and Tips for Succeeding in Today's Business World
On the American side, the White House published in 2024 application guides for the Executive Order on AI, reinforcing transparency and security testing requirements for large models, particularly in health, finance, and infrastructure. The approach remains sectoral, creating a patchwork of obligations depending on the field of activity.
China, for its part, has been enforcing its rules on generative artificial intelligence services more stringently since 2024: model registration, prior control of training data, content censorship mechanisms. For companies operating in multiple markets, we observe that triple compliance is becoming a full-fledged architectural consideration, not just a legal issue. Those following Bozar’s tech articles will regularly find analyses on these intersecting constraints.
Further reading : The latest trends and must-know news in the automotive world in 2024
AI PC and NPU: AI Computation Migrates to the Endpoint
The year 2024 marks the concrete arrival of “AI PCs” equipped with integrated NPU (Neural Processing Unit) in the SoC. Intel, AMD, and Qualcomm now offer chips capable of executing inference tasks locally that, just a few months earlier, required a cloud call.
This shift is not cosmetic. It changes the data processing chain for enterprise applications:
- Sensitive data preprocessing (voice recognition, document analysis) remains on the device, reducing network exposure and latency issues
- Compact language models, optimized to run on NPU with a few dozen TOPS, are sufficient for summarization, classification, or contextual suggestion tasks
- Energy consumption per inference drops compared to a remote API call, impacting the carbon footprint of large-scale terminal fleets
The workstation becomes an active AI computation node, not just a thin client. For CIOs, this means rethinking hardware provisioning and renewal cycles by integrating NPU capacity as a selection criterion.
Post-Quantum Cryptography: A Technical Migration Already Underway
The quantum threat to current encryption algorithms (RSA, ECC) is no longer theoretical enough to be ignored. NIST finalized its first post-quantum cryptography standards in 2024, and major cloud providers are already integrating these algorithms into their TLS layers and key management services.
The problem for companies is not the algorithm itself, but the inventory. Identifying all points in the information system where vulnerable encryption is used (certificates, VPNs, code signatures, encrypted storage) represents a considerable undertaking. We recommend starting with the flows most exposed to “harvest now, decrypt later” attacks, where an adversary stores encrypted data today to decrypt it later with a quantum computer.
The migration will not happen in a single budget cycle. Companies that delay auditing their cryptographic inventory accumulate a silent technical debt.
Generative AI in Production: From Prototype to Industrial Pipeline
The majority of companies experimented with generative AI in 2023. In 2024, the focus has shifted to industrialization. The difference lies in three technical axes that mainstream overviews rarely address.
The first concerns RAG (Retrieval-Augmented Generation). Connecting a language model to an internal document base via a vector index has become the standard pattern to reduce hallucinations and anchor responses in verified data. The quality of chunking and embedding directly conditions the relevance of the results.
The second axis is observability. Monitoring a generative model in production does not resemble traditional application monitoring. It is necessary to trace the quality of responses, detect behavioral drifts, and measure inference latency, all without slowing down the pipeline.
The third is cost. Cost optimization per token becomes a strategic lever as soon as the volume of calls exceeds a few thousand per day. The choice between a proprietary model via API and an open-source model hosted internally depends on the balance between inference volume, data sensitivity, and available GPU capacity.
Sustainable Technologies and Digital Sobriety: An Architectural Criterion
GreenTech is no longer a marketing argument. European regulatory constraints on non-financial reporting (CSRD) push technical departments to integrate energy consumption as a design metric, alongside performance or availability.
Data centers are adopting liquid cooling strategies and recovering waste heat. On the software side, “green coding” practices aim to reduce the footprint per request: optimizing database queries, compressing AI models, choosing event-driven architectures over constant polling.
Digital sobriety becomes a technical arbitration parameter, not just a pious wish in a CSR report. Teams that genuinely measure their consumption per feature gain a concrete advantage over those that merely offset.
The technological trends of 2024 share a common trait: they shift complexity. AI regulation creates new layers of compliance engineering. NPUs redistribute computation between cloud and endpoint. Post-quantum cryptography imposes an inventory that no one had planned. Generative AI in production demands operational skills that did not exist two years ago. Each of these challenges rewards teams that tackle them early.