Windows 的过去、现在和未来

Phi-Silica失宠了

不过怪不得这么快，CEO指定的事情

SATYA NADELLA: …And they will soon be able to run DeepSeek’s R1 distilled models locally on Copilot+ PCs, as well as the vast ecosystem of GPUs available on Windows…

何木槿 好像没什么问题……云是核心，那么终端就是边缘了

DaleZ 我觉得不太会，至少纳德拉在刚才的电话财报会议中不是这么表示的。

KARL KEIRSTEAD, UBS: Thank you. Maybe this one as well for Satya, and it’s also away from the numbers. But, Satya, I wanted to ask you about the Stargate news and the announced changes in the OpenAI relationship last week. It seems that most of your investors have interpreted this as Microsoft, for sure, remaining very committed to OpenAI’s success, but electing to take more of a back seat in terms of funding OpenAI’s future training CapEx needs.
I was hoping you might frame your strategic decision here around Stargate, and Amy, whether there’s any takeaway for investors from that decision in terms of how you’re thinking about CapEx needs over the next several years. Thank you.

SATYA NADELLA: Yeah, thanks for the question. We remain very happy with the partnership with OpenAI. And as you saw, they’ve committed in a big way to Azure. And even in the bookings, what we recognized is just the first tranche of it. And so, you’ll see, we’ve given the role for we have more benefits of that even into the future. And obviously, their success is our success because even all the other commercial arrangements that we detailed out in the blog that we put out even commensurate with that announcement.
But to your overall point, the thing that I would say is we are building a pretty fungible fleet. We’re making sure that there’s the right balance between training and inference. It’s geo-distributed. We are working super hard on all the software optimizations, I mean, just not the software optimizations that come because of what DeepSeek has done, but all the work we have done to, for example, reduce the prices of GPT models over the years in partnership with OpenAI. In fact, we did a lot of the work on the inference optimizations on it, and that’s been key to driving.
One of the key things to note in AI is you just don’t launch the frontier model, but if it’s too expensive to serve, it’s no good. It won’t generate any demand. You’ve got to have that optimization so that inferencing costs are coming down and they can be consumed broadly.
And so, that’s the fleet physics we are managing. And also, remember, you don’t want to buy too much of anything at one time because of the Moore’s Law every year is going to give you 2x. Your optimization is going to give you 10x. You want to continuously upgrade the fleet, modernize the fleet, age the fleet, and at the end of the day, have the right ratio of modernization and demand-driven monetization to what you think of as the training expense.
I feel very good about the investment we are making. And it’s fungible, and it just allows us to scale more long-term business.

何木槿 betaflt能在Dev停9个月，SV2 Moments能蹭上Current工程续命一年多，魔幻的WIP

ThinkBou 以ONNX QDQ格式优化的DeepSeek R1 7B和14B蒸馏模型已经上线VS Code AI Toolkit，同1.5B模型一样率先登陆搭载Qualcomm Snapdragon X的Copilot+ PC。

可用于本地NPU运行。14B模型速度约为8 T/s，1.5B模型速度约为40 T/s。