xadhominemx

11 ms·

31.
by xadhominemx 1y ago
There are workarounds and paths forward, but they are definitely being slowed down. I know a lot about this topic.
32.
by xadhominemx 1y ago
That’s just because they spun off all the high quality companies (Agilent, Keysight, Verigy, Avago). The PC server and consumer print business have always been commodity product.
33.
by xadhominemx 1y ago
The US is certainly slowing down China considerably. China would certainly not have an import ban on Blackwell GPUs if they were made available. And upstream, the ban on EUV and other high end semiconductor production equipment has severely
34.
by xadhominemx 1y ago
I’m sure all their infrastructure and probably the majority of their employees will be in the NYC metro area.
35.
by xadhominemx 1y ago
They are doing it for the love of the game, IMO
36.
by xadhominemx 1y ago
Consensus EPS for FY27 (~CY26) is $6.40. Buy side is higher!
37.
by xadhominemx 1y ago
If we are not headed to ASI, the spending will slow down and the problem will solve itself.
38.
by xadhominemx 1y ago
Their earnings will certainly decline or at least decelerate if capex slows. I’m just saying, if the market wasn’t pricing in a slowdown, NVDA would be trading at 40-60x next year EPS, not 25x.
39.
by xadhominemx 1y ago
They just committed to invest $100b (!) in OpenAI and said $100b is only the start.
40.
by xadhominemx 1y ago
If we get ASI it will figure out how to do the green transition for us!
41.
by xadhominemx 1y ago
Hyperscalers are only spending less than half of their operating cash flows on AI capex. Full commitment to achieving AGI within a few years would look much different.
42.
by xadhominemx 1y ago
NVDA stock does not trade at a huge multiple. Only 25x EPS despite very rapid top line growth and a dominant position at the eve of possibly the most important technology transition in the history of humankind. The market is (and has been)
43.
by xadhominemx 1y ago
More efficient inference = more reasoning token. Hyperscaler ASICs are closing the gap at the hardware/system level, yes.
44.
by xadhominemx 1y ago
Test time compute has made consumption highly elastic. More compute = better results. Marginal cost of running these GPUs when they would otherwise be idle is relatively very low. It will be utilized.
45.
by xadhominemx 1y ago
For large models, the bottlenecks are memory bandwidth, network, and power consumption by the DAC/ADC arrays It’s never come even close to penciling out in practice. For small models there are people working on this implemented in flas
46.
by xadhominemx 1y ago
I understand that topic well. They stitched top metal layers across the reticle - not that challenging, and the foundational IP is not their own. Everyone else went the CoWoS direction, which enables heterogeneous integration and much more
47.
by xadhominemx 1y ago
Cerebras hasn’t made any technical breakthroughs, they are just putting everything in SRAM. It’s a brute force approach to get very high inference throughput but comes at extremely high cost per token per second and is not useful for batche
48.
by xadhominemx 1y ago
I don’t think so. The reason why Cerebras is so fast for inference is that the KV cache sits in the SRAM.
49.
by xadhominemx 1y ago
Why would you say 16A would never be manufactured in the United States? It is of course TSMC’s plan of record.
50.
by xadhominemx 1y ago
In 24 months, US hyperscalers will be training models on GPUs/XPUs with 16A process technology and HBM4E. The gap between the raw processing power of US and Chinese AI hardware will be widening.
51.
by xadhominemx 1y ago
“A modular iphone that has an easy to replace battery, easy to replace screen and is maybe 2mm thicker to account for it?” Definitely not! This would be an inferior product in almost every respect for 95% of customers.
52.
by xadhominemx 1y ago
There are visa types specifically created for this sort of situation (e.g. E2 visas). But those visas are only available to certain nationals and South Koreans are not among them, which is very stupid given the strong commercial and strateg
53.
by xadhominemx 1y ago
It’s not a language issue - it’s because a quirk in US immigration law means South Koreans are not eligible for the E-2 visa, and are stuck competing for capped categories like the H-1B. A group of house reps have been trying to fix this is
54.
by xadhominemx 1y ago
Extremely large and fragmented community of AI researchers and developers in China. CUDA competitors in China face the same hurdles as those in the West.
55.
by xadhominemx 1y ago
It’s because the chips are networked with very high bitrates and need to be physically densely packed together.
56.
by xadhominemx 1y ago
Exactly, 4 blower GPUs!
57.
by xadhominemx 1y ago
Oof I don’t know about that. Driving in NYC is much different than San Francisco. Frequent lane departures, cutting into heavy traffic despite technically lacking right of way, and other moderate rule breaking is required to get anywhere. B
58.
by xadhominemx 1y ago
> My experience is that gpt-oss doesn't know much about obscure topics That is the point of these small models. Remove the bloat of obscure information (address that with RAG), leaving behind a core “reasoning” skeleton.
59.
by xadhominemx 1y ago
The ICs are tiny but the antenna are much larger.
60.
by xadhominemx 1y ago
The chip is still the majority of the cost of the inlay, so making a smaller chip is the key to driving total down and unlocking new use cases.

More ›