Nvidia dominance challenged by Vera, neo-clouds, ARM & custom silicon
Key Questions
How is Microsoft reducing reliance on OpenAI?
Microsoft's AI chief stated the company was 'set free' from OpenAI to pursue superintelligence. It is developing in-house models like MAI-Thinking-1 and the Maia 200 chip.
What is NVIDIA's RTX Spark superchip?
NVIDIA's RTX Spark combines Arm and Blackwell 3nm tech for consumer AI PCs. It brings AI supercomputing capabilities to edge devices and personal computers.
How is Cerebras challenging NVIDIA's dominance?
Cerebras refuses to work with NVIDIA and promotes its WSE-3 wafer-scale engine as an alternative. It targets large model training without GPU dependencies.
What alternatives to NVIDIA GPUs are gaining traction?
Tenstorrent's open-source RISC-V with GDDR6 offers cost-effective inference, while Arm sees $2B order visibility for CPU-centric agentic workloads. Intel's Crescent Island uses LPDDR5X.
How is Broadcom's AI chip outlook affecting the market?
Broadcom missed its AI chip forecast at $56B versus $57.6B expected, signaling demand softening. Its CEO noted the custom chip race is fragmenting beyond NVIDIA.
What is NVIDIA's Vera Rubin NVL72 designed for?
Vera Rubin NVL72 incorporates CPU orchestration with Olympus cores and FP8.2 to address GPU feeding bottlenecks. It strengthens NVIDIA's data center stack for enterprise AI.
How are European and Chinese firms competing in AI chips?
Mistral is considering in-house chips for European sovereignty, while DeepSeek's $7.4B round with Tencent challenges the 'more chips' narrative. Seres shifted to ByteDance's NVIDIA-backed AI.
What networking advancements address AI bandwidth issues?
Cisco Silicon One and Google’s Marvell TPU networking tackle connectivity walls. These support distributed inference as agentic AI scales beyond single-vendor ecosystems.
Intel Crescent Island LPDDR5X alternative; Nvidia acquires Kumo AI; Europe's push for homegrown chips; Microsoft's own models reduce reliance; Nvidia RTX Spark (Arm+Blackwell 3nm); Seres ditching Huawei for ByteDance's Nvidia-backed AI; DeepSeek's $7.4B funding; Broadcom AI chip forecast miss; Cerebras refuses to work with Nvidia; Spirit AI tops Nvidia in physical AI ranking; NetApp/Cisco FlexPod for AI uses NVIDIA; Jensen Huang COMPUTEX keynote; DeepSeek V4 100x cost advantage; Google taps Marvell for TPU networking; Cerebras WSE-3; Meta halts custom chip deal with Samsung; Nvidia releases Nemotron 3 Ultra; Tenstorrent's open-source RISC-V; FriendliAI offers Nemotron 3 Ultra; Microsoft's Maia 200 chip and MAI model; Nvidia Vera Rubin NVL72; Nvidia picks Unitree for GR00T; Broadcom's AI reality check; Broadcom CEO confirms custom AI chip race fragmenting; Jensen Huang portfolio analysis; Netrasemi A2000 edge AI chip; Hitachi-Intel physical AI collaboration; Microsoft AI chief declares independence; Mistral considers in-house chips; Cisco Silicon One; Arm's $2B order visibility. New: Micron joins Nvidia's HBM4 supplier lineup for Vera Rubin, confirming three-way race (Micron, Samsung, SK Hynix).