witeken

[exclusive] Nvidia to buy Intel! To become the world's premiere AI IDM. Leaked footage shows Jensen Huang discussing with Pat Gelsinger about the acquisition. (Nvidia is rumored to pay about $69.42 per share.) https://t.co/Ke3UFMyF3h

Breaking news: Nvidia to acquire Intel in all-stock transaction! Pat Gelsinger: we've failed miserably in AI. With this merger, we’re suddenly #1 market share. Jensen Huang: Intel is so cheap (~1% dilution) we couldn’t overlook adding their CPU and manufacturing capabilities.

Exactly why the whole CUDA moat story is nonsense: https://t.co/JXz1x4wKOV

Intel & AMD are significantly cheaper, which is basically the same. It's not Hopper/Blackwell vs. MI300/Gaudi3. It's 1 Hopper vs. 2 MI300s/Gaudi3s or 1 Blackwell vs. 3 MI300s/Gaudi3s. E.g. H100 is already significantly slower than 1 Gaudi3, so never mind 2 Gaudi3s (same price). https://t.co/avj2QdLKiv

Delusional post; dismissing Gaudi3 is just nonsensical. It is faster than Hopper, and by quite a lot (40-100%). _And_ it already costs 50% less. So it has 3-4x higher performance per dollar. Looking at the tens of billions of Nvidia revenue, clearly sw/dev cost is negligible. https://t.co/JdzVy7HyAZ

AMD (Q1) net profit margin: 1% (21% non-GAAP) Nvidia (Q1) net profit margin: 57% So Nvidia will not only have ~5x higher revenue in 2024 ($24B vs. $120B), but also (depending on the accounting scheme) 2.5-50x higher profitability. https://t.co/0CR8bHs3Le

But Nvidia won't. https://t.co/E5Yx7loiLX

Indeed good luck to Nvidia competing against 18A(-P)/14A Intel (RibbonFET/PowerVia) products with deprecated 3nm FinFET technology into 2027. https://t.co/imUUgFYARS

The vast majority of that (besides Moore's Law) has been due to the introduction of fixed function Tensor Cores, Transformation Engine, lower precision and sparsity. None of that is exclusive to Nvidia. It's like if Intel had gradually introduced GPU hardware into its CPUs. https://t.co/UUQ4RWiOxx

Indeed good luck to Nvidia for having to compete against Falcon "angstrom-era process 1500W TDP" Shores with 5nm-class Blackwell Ultra. Nvidia get rekt into oblivion. https://t.co/imUUgFYARS