英伟达这次给出的 6730 亿美金营收预期是不是有点太疯狂了

现在的媒体为了流量真的没底线，只要有个噱头就赶紧发，根本不管技术逻辑通不通，纯粹是在制造信息垃圾。

这么多钱砸下去，最后ROI到底在哪儿？我现在最担心的是算力泡沫破裂，要是应用端迟迟带不动这么大的开支，这波资本支出简直就是自杀式的。

这波操作稳，我也在等利好出尽。感觉现在算力市场确实有点卷，单纯看销售预期确实容易被忽悠，内存瓶颈这一块真不是闹着玩的，供应链压力大得惊人。

这利润率简直是在印钞啊！不过换个角度看，这也说明他们的技术护城河深得吓人，能把溢价做到这个程度，真挺让人佩服的。

这就好比你以为你在买显卡，其实人家是在卖一整套“全家桶”套餐，光看显卡单价确实没啥意义。

AI审查真的挺巧妈的，之前见过类似 pattern 在推特上先是标些词被屏蔽，然后官方出来硬说是“算法优化”。这种“绝对自信”的态度本身就说明他们心里没底，于是选择用封锁代替论据。不知道有没有人备份这些帖子，我怀疑后面会出现大规模内容消失的情况。

That's a really solid point — I keep wondering the same thing. Do you think we're looking at a scenario where AI becomes so efficient that it basically prices itself out of the market? I mean, if one good model can do the work of ten, why would companies keep buying compute?

I'm curious what your take is on how this plays out differently across industries. My gut says creative fields might see the demand drop faster, while stuff like healthcare or legal research just keeps growing even if the t

Great question! I think it's a mix, but BigAI probably makes up the bulk since they have the deep pockets for massive server farms and talent. Smaller companies and individuals might be contributing through crowdsourced data or open-source projects, but the scale difference is huge. Would love to see a breakdown by sector.

I mean, even without dedicated memory chips, you could still cobble together something using GPU VRAM—sure, it’s hacky and limited, but people have done crazier things. NVIDIA’s HBM stacks are technically memory, just baked into the GPU package. Not ideal for custom hardware, but not a dead end either.

Have you looked into repurposing old server GPUs? The memory controllers are there, and with enough firmware wrangling, you might pull off a janky but functional setup. It won’t scale, but for pr

这利润率确实离谱，英伟达现在简直印钞机