硅谷现在最怕的不是中国AI追上来了
是中国AI追上来的方式跟比亚迪一模一样
性能不输你,价格把你打穿
然后还开源让你自己拿去免费用
上周月之案面发布的Kimi K3就是这剧本的第三季
第一季是Deep Sea
第二季是通议签问
现在轮到Kimi了
美国一众分析师在节目里直接搬出了比亚迪的历史
当年底特律也是这么笑比亚迪的
现在全球每三辆电动车就有一辆是比亚迪的
中国AI太可怕了
也许是时候封禁了
来看美媒是怎么报道的
也有想到的想法,但这个是 Moonshot,也许是什么新的,也许是新的。
我许是中美国的特别人,也许是中美国的特别人,也许是中美国的特别人,也许是中美国的特别人,也许是中美国的特别人。
这一种是一种的 surprise,我们看到了解,我们看到了一些新的特别人,也许是中美国的特别人,
包括 Alibaba,一个公司的公司,也许是中美国的特别人,也许是中美国的特别人。
但在那一天,我們有過這次的新聞,
他們推出了金米K3,
一種非常強烈的模式,
它外出了很多的 Frontier models
在很多的目標,
除了兩位主持人,
從 Anthropik和 OpenAI,
所以它是非常強烈的模式,
它有2.8 trillion的模式,
是相當的相似的
brain synapses,
它是非常強烈的模式,
我認為它會讓人有一個新的模式,
那是有一個新的模式,
that was able to come and compete with the best from Silicon Valley
on a number of parameters at cost that people really don't understand at this point.
Many of these Chinese models are also open source or open weight,
so they're much less expensive.
They're offered very broadly.
If you don't want to run them on somebody else's computer systems,
you can run them on your own.
Also, you can take the model and then run it,
so that can avoid some of the security concerns around Chinese models.
but it is a very serious threat for some of these frontier models in the U.S.
that had been planning on charging premium prices.
Now there may be some pricing pressure on what they're able to charge their customers,
and that may make it a little bit more difficult as we see OpenAI and Anthropic
try to go public in the near future.
Both of them want to get valuations that are above a trillion dollars each.
That may be a little bit more challenging today.
Right, and it challenges the whole CapEx thesis and the whole AI stack.
Peter,Isabel made a really good point earlier when we were talking on our pre-show call
basically comparing this to BYD
I mean BYD,you have this Chinese technology
that's just as good as not better than other EVs in the US
and it's cheaper,but notably in America
you can't buy a BYD
Do we hear any noise coming from government officials
about trying to curtail the usage of some of these Chinese models?
There are questions about this
I mean you make a very good point
China has been able to compete in a number of different areas
not just electric vehicles but also solar panels
steel for example
so you do have some of these historic cases
where in different industries Chinese companies have come out
they've been able to do things at much much lower prices
yes that is happening in AI models right now
and you've got a whole series of them
DeepSeek, Moonshot now
there's Minimax
there are others that are coming out with these less expensive models
and there has been some discussion about whether we should try to
bar those companies from competing in the US
I think the challenge there is that you may be hurting American companies
by forcing them to use more expensive models or less capable models
there are open source models in the U.S. too
but I think when you look at the leading edge of open source right now
these less expensive models
they're really Chinese companies at this point
so if you take away a tool that companies want to use
to be able to incorporate AI into their businesses
you may end up hurting American businesses that are not the AI leader
所以你可能会帮助了AI的中央企业企业,但你可能会罵其他企业企业企业的社会。
Peter,所以 interesting.
Remember, chip stocks fell into a bear market Friday,
在 part因为的问题是由于咨询的KEMEK3,中国的MoonShot。
我们的下来是正常的,但就像100,加上100,加上100,加上100,加上2%,
and it's a lot of the compute names nvidia but also memory names that are driving some upside in those public markets
since moonshot released this can be k3 model last week everyone's debating what it really means
does more efficient ai mean less hardware is china closing the gap faster than we expected
where does ai's real value ultimately accrue joining us now is crosslink capital partner
Mary Donofrio
focused on venture growth investments
but across software, AI
and infrastructure
and for the last three days
I've said Kimi K3 many times
but it was interesting to like
take a sort of straw poll of
VCs, founders
people in industry of how they
react to it so differently
some people are very fearful, defensive
others say this is very good
for AI as a whole
very difficult job first thing on a Monday
but your reaction to what happened
Kimi K3, as you said, was phenomenal
and it's similar to the DeepSeek moment we had last year.
2.8 trillion parameter model
and what it demonstrates, I think,
is that China and open-weight models
are closing the gap with frontier models.
Obviously, it was done based on their own benchmarking
but according to the benchmarks, they're fourth.
And while you can debate the merits
of those benchmarks themselves,
the productivity gains are tremendous
and I think that in general it is part of the pathway towards
more widespread open weight model usage
and in the context of enterprise budgets ballooning in AI
I actually think that it's probably a boon for AI ubiquity over time
when you think about your existing portfolio of companies
were you sort of like okay how does this impact their trajectory
has anything actually fundamentally changed for the
the different pieces of the AI5 layer cake
that you're invested in?
I don't know if it changes the cake itself,
but I do think it's another demonstration
that models keep leapfrogging one another,
and if companies don't out-innovate,
they're going to be leapfrogged too.
Whether it's,you know,
every single model seems to outperform the last,
and it's a matter of stay out-innovating
by creating modes of data,
of performance,
of customer resonance,
另外, unfortunately, the innovations in both Frontier and Openweight models
are tremendous.
It's not a sophisticated argument, but when this all happened,
a lot of people came out and said,
well, if a $20 billion Chinese startup can do a $2.8 trillion parameter model
with these economics,
why are we assigning OpenAI and Anthropic evaluation of a trillion?
They looked at the sort of fundamentals of that.
是那是正確的方法想想想?
frankly
I don't know yet and I don't think
anybody knows yet because even
Because even in the context of those questions,
Anthropic and OpenAI are still producing billions and billions of revenue.
And it's not just from their, you know, they have first-party businesses,
they have third-party businesses.
and I think, you know, obviously the API opened for Moonshot,
but the open waits have not yet.
Yeah, 27th of July.
And it's worth noting that because it's open wait doesn't mean it's free to run.
And so I think that's a dynamic to these models
that people potentially underappreciate as well.
So much back and forth, Kate, for today's tech check.
Kelly, so much to cover.
A lot of it has to do with China.
You mentioned Alibaba taking the spotlight this week
with its preview of a model that it claims now is a runner-up to rival Anthropics.
best version out there. These are internal benchmarks so far, so we'll see where they
stack up in the coming weeks. It does come after Moonshot's Kimi AI model took the tech world by
storm just last week. According to independent benchmarks, it is almost on par with Anthropics
and OpenAI's best versions out there. It is also the world's largest open weight model.
Did see a lot of demand. The company's saying over the weekend that they are pausing
subscriptions. That goes back to the compute conversation. CEOs I'm talking to, though,
really are framing this new generation out of China as a capability shock. The takeaway has been
it is no longer a given that the U.S. frontier labs have a safe lead in AI. It is starting to
show up in some of the data. Open router, for example, showing Chinese models overtook U.S.
usage last week, if you measure that by token usage. This could be a lot of experiment experimenting
out there, Kelly, but we'll see where that does net out. Open AI chairman Brett Taylor
on Squawk Box earlier did pour a little bit of cold water on the idea that this all just comes
down to cost. He said it they may be a little less efficient. I think it's a little bit overblown
about these open weight models is they're not necessarily cheaper to run. Whether or not they're
cheaper to train, you don't care. All of this, Kelly, is spilling over to geopolitics as well.
A report from Axios this morning says the U.S. is now considering banning some of these cutting
edge AI models out of China. That got a lot of buzz over the weekend on Twitter, Kelly.
Kimmy K3 this time is not a fan, it's a開源
Chinese companies over to be mad just will抄襲
After they were mad just will get a closed system
Now they have been a closed system
All the world is open for money
You can download, you can't do it
You can't even get it
Because you will be a cop
You can't do it
You can't do it
You can't do it
You can't do it
You can't do it
人家说那好吧我开源给你看
这个剧本从来没在硅谷的应急手册里出现过
硅谷过去三年一直在内部辩论同一个问题
开源会不会吃掉必源的午餐
OpenAI说不会
Anthropic说不会
因为他们相信最好的模型永远需要最昂贵的训练
然后Deepseek V3开了
综合基准测试跟GPT4O几乎拉不开差距
训练成本据估算只有500多万美元
是 GPT-4O的一个零头
然后是阿里的通议签问困2.5
开源权重在多个内部基准上已经追到Cloud Sonny身后一个身位的位置
全球开发者下载量冲到模型排行榜第二
现在MoonShot也开了
Kimmy K328000亿参数
独立基准跟Cloud的最强版只差一两个百分点
三个中国模型排成一排
每一个都踩在同一条路径上
用远低于硅谷的成本造出几乎一样好的东西
然后开源让你自己去验证
这已经不是单个公司的事了
这是中国AI产业的结构性产出
这三个例子里
每一个开源的逻辑各不相同
这才是最让硅谷难受的地方
Deepseek开的核心是性价比
他用混合专家架构
在训练阶段就压了成本
然后用开源把成本优势
分享给所有人
等于是说
我花了500万训的东西
你免费拿去跑
你花5000亿训的东西
凭什么让我付钱
Ali的开源逻辑是生态所
困开源后
全球开发者
基于他做了成千上万个
定制版本和微调应用
这些应用在回流到
Ali云的模型服务生态里
不管你跑谁的版本
里程算力总有一部分
落在 Ali云上
开源不吃你的授权费
但要你的云账户
Moonshot的开源逻辑更直接
公司估值才20亿美元
没有万亿包袱
不需要填估值故事里的营收窟窿
干脆把模型扔出去
让全球开发者自己玩
口碑传播本身就是最好的融资材料
三加三个打法
共同效果是
全球开源模型的领先位置被中国公司坐稳了
CNBC采访的那个分析师说
Kimi K3是开放权重模型
不代表免费运行
还有推理费用
这是在偷换概念
要付推理费这件事
是所有模型的共同前提
你跑Claw的也得给云厂商付钱
跑Kimi K3也得付
云费是所有玩家平摊的基础设施成本,
矽谷前沿实验室能拿来填万亿估值的,
从来不是云费,是模型授权费,
是API调用抽成,
现在开源把模型本身的定价踩到零,
你凭什么收纳成溢价?
这才是从根上炸掉了估值模型。
中国公司把开源当成战略武器在用,
这不是一句比喻,
美国用芯片封锁算力供应,
试图卡住中国AI的喉咙,
中国用开源锁住你的商业模式
让你的定价权自焚
你没办法进开源
开源不经过海关
不需要签证
不需要任何人的批准
它是互联网的原始协议
是http和tcp的同一种逻辑
你不可能通过立法禁止http
硅谷这次被自己开创的互联网底层规则反噬了
如果要把时间线拉长来看
这个局面三年前就已经埋下了
2022年美国开始限制高端GPU出口中国
硅谷当时的主流判断是中国AI很快就会被卡死,算法再强你也得跑在硬件上,硬件卡住你,算法就是纸上谈兵。
三年过去了,回头看那个判断,它只对了一半,硬件确实被卡住了,中国公司拿不到最先进的英伟达旗舰卡。
But the way of the AI team is being pushed to the entire card.
The way of the AI team is being pushed to the架限,
the environment, the environment, the heat of the energy,
the air flow, the air flow, the air flow, the air flow,
all the way of the AI team is being pushed to the entire card.
Because you don't have a card, you must be able to give the card to your
the way you do it.
Your opponent is not able to get the card,
you don't have to pay, you don't have to pay,
you don't have to pay, you don't have to pay,
It was a good thing to do with the fact that it was a good thing.
This is the story of the day of the航空.
The American in 2011, through the Wolf-Torchel,
禁止 NASA and the航空船 to work,
trying to put the country in the international space.
After 10 years,
the country has been created by its own space.
The sky is from the sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky sky
On the top of the page, it just wrote a few words,
don't buy them, just use them, look who better.
20 million dollars of the year's account,
wrote this letter,
on a million dollars of the year's account,
they didn't have to do this letter,
they didn't have to do it,
because any of the year's year's account,
they didn't want to open up their own way,
they didn't want to open up any money,
but you don't open up,
the Chinese companies will be able to open,
。