Nvidia invests $3.5 billion in MediaTek AI chip link
MediaTek will adopt Nvidia’s NVLink Fusion platform for custom AI accelerators tied to Nvidia rack-scale systems.
Chips, lithography, efficiency, quantum: more computing for fewer watts, and what we will do with it.
Computing is the hidden resource behind everything else: without efficient chips there is no artificial intelligence, no autonomous vehicle, no smart grid. The section follows process nodes shrinking — 2 nanometres and beyond — architectures doing more operations per watt, memory, interconnects, and quantum computing whenever it produces verifiable results rather than promises.
We read the roadmaps of TSMC, Samsung, Intel and the Chinese foundries, the architecture papers, the independent benchmarks. The angle never changes: efficiency — compute per unit of energy — matters more than raw performance, because it decides what will actually run, from the data centre to the phone. Every article gives its orders of magnitude and cites its sources.
MediaTek will adopt Nvidia’s NVLink Fusion platform for custom AI accelerators tied to Nvidia rack-scale systems.
CUDA Python 1.0 bundles several packages under a shared versioning strategy.
Tests on Semianalysis’s InferenceX benchmark put Jalapeño ahead of Nvidia Blackwell on two efficiency measures.
New funding lifts Starcloud's total raised to $450M and its valuation to $2.3B.
Six large-model training workloads each exceeded 98% of NVIDIA reference performance.