--- author: Joseph Wang date: '2026-10-10' guest: '' layout: post.njk source: https://www.youtube.com/watch?v=so2GEr0oqSg speaker: Joseph Wang tags: - treasury-yields - inflation-expectations - monetary-policy - ai-capex - energy-refining title: 美债收益率筑顶、能源通胀约束与 AI 投资回报率的现实拷问 summary: 本期分析了美债收益率在强劲拍卖与高企 TIPS 实际收益率支撑下的企稳信号，探讨了美联储加息预期转向及中东地缘政治和精炼油短缺对通胀的压制，并深入剖析了 AI 资本开支激增与企业实际营收增速脱节带来的估值脆弱性与电力物理瓶颈。 insight: '' draft: true series: '' category: investment-assets area: finance-wealth project: [] people: - Christopher Waller - Steve Hsu companies_orgs: - OpenAI - Anthropic - Federal Reserve products_models: [] media_books: [] status: evergreen --- ### 美债拍卖强劲与通胀保值债券的历史性吸引力 上周美股标普 500 与纳斯达克指数创出历史新高，显示市场上升趋势依旧明确；与此同时，此前持续攀升的债券市场迎来了关键的企稳信号。过去数周，**10 年期美债收益率**（10-Year Treasury Yield: 衡量无风险借贷成本与长期通胀预期的基准利率）几乎每天都在被单边抛售，引发市场对 1987 年式股债崩盘的担忧。然而，最新一轮 10 年期国债拍卖成功遏制了这一势头。在衡量国债拍卖健康度时，市场主要关注中标收益率相对发行前交易预期的利差（即通过 When Issued 产品衡量的借贷意向）、**投标倍数**（Bid-to-Cover Ratio: 投标总量与实际发行额的比值，代表市场真实认购意愿）以及**一级交易商获配比例**（Primary Dealer Takedown: 法律强制承销商兜底的被动承接份额）。此次拍卖中标利率低于预期，且伴随着极高的认购倍数和极低的一级交易商接盘比例，表明私营资本在当前利率水平下具备极高买盘，为长期收益率设定了阶段性软顶。 在名义国债之外，**通胀保值债券**（TIPS / Treasury Inflation-Protected Securities: 本金随 CPI 调整以对冲通胀的国债品种）展现出数十年罕见的吸引力。目前 30 年期 TIPS 提供的**实际收益率**（Real Yield: 扣除通胀预期后的无风险真实投资报酬率）高达约 3.3%，意味着投资者在获得固定票息的同时，本金还将每年依据 CPI 全额上调。部分市场人士怀疑 CPI 的真实度，但从制度经济学角度来看，通胀本质上是一篮子商品权重加权的抽象统计工具，并不存在统一绝对的“真实通胀”。无论在哪种经济体，官方 CPI 都能较好反映宏观价格变动趋势，足以实现跨周期的财富保值。结合名义与实际收益率水平，固定收益资产正迎来结构性长期资金的承接。
Original English Hello my friends. Today is October 10th and this is Markets Weekly. All right, guys. So, last week was a good week in the markets. We saw the S&P 500 and the NASDAQ make new all-time highs. That tells me the trend is still higher. In addition, we had stability in the bond market, which is of course a welcome reprise. So, today let's talk about two things. First off, let's talk about what's causing the stability in the bond market. And secondly, let's talk about the little scare in the AI trade we got due to reporting from the FT on revenues of Open AI. All right, starting with the bond market. So if you look at the 10-year yield, you can see that over the past few weeks, yield has gone up relentlessly. It seems like every day we wake up and someone sells bonds. So many people were wondering just how does this end, right? Are we just going to go up and up and up? Historically we have seen the bond market crash like in 1987 and that did also bring down the equity market in a big way as well. The good news is last week we seem to have reached a level of yields that is drawing a lot of investor interest. So it seems like the stability in the 10-year yield was precipitated by a very good 10-year auction. So one of the ways that people can see whether or not there's investor demand is to see what happens at the auction. That's when of course the government has to sell debt to the market. Now people evaluate the success of an auction based on a number of metrics. A very common one is to see where the yield cleared relative to market expectations which are formed through a futures product called when issued. So last week we did see the auction clear lower than what was implied by the when issue. That is to say through it. So that was a good sign. In addition, people also look at other metrics such as bid to cover and just how much did the primary dealers take down. Now bid to cover is basically how much demand there was relative to how much was on offer. And primary dealer takedown: basically the primary dealers are obligated to bid at auction and so you can never have a failed auction, right? So like anyone who tells you that the auction can fail, that's just doomer nonsense. The auction can never fail. All the primary dealers are obligated by law to bid on it. Now what happens is that if there's not a lot of demand from the private sector then the primary dealers take up a larger share of the auction and then they have to warehouse that and then sell it off later on. So what we saw from the auction metrics is very strong bid to cover and also very low primary dealer takedown and that tells everyone that there is a lot of demand for 10-year notes at this level of interest rates. So that kind of puts on a soft ceiling for yields for the moment and tells the market that there's demand here and maybe we won't have to wake up every day and sell treasuries. In addition, I'll also note that if you look at inflation protected treasuries, they are historically very attractive. So 30-year TIPS are trading about 3.3% yield and that's at multi-decade highs. And the way that this product works is that say at this 3.3% you would get a 3.3% coupon in addition your principal would adjust higher by let's say inflation by CPI every year. So let's say if CPI was 3% your principal would go up by 3%. So that is basically why it's called inflation protected, right? Your principal goes up with CPI and in addition you still get that 3.3% yield which of course is why it's called a real yield. So if you're someone with a lot of money or just really worried about inflation in the future, this is a really good way to preserve wealth. And of course if the market implied say break even is like say 2.2%, if CPI were to come in higher than that you would be doing better than nominal treasuries. So many people would also say that oh you know you can't trust the CPI blah blah blah, but I think that really misunderstands this whole thing, right? So first off there's no such thing as true inflation, inflation is a social construct, right? So think about how you make an inflation index: you have thousands of items, each of them go up and down in price to different extents. You could have RAM going up 400%, you could have gasoline going up 20%, or you could have some other product that's actually only going up 1%. So you have all these products that go up and down every year in price at different extents. So you include them in the basket and then you have to decide what to include and just as importantly you have to decide how much to weigh each product. Do you weigh say RAM very heavily or do you weigh milk heavily and so forth. So it's all just an abstraction. It's basically something that people come up with as a very rough measure of the overall price level. It's going to be wrong for basically everyone, right? For some people it would be too high, some people it would be too low. So to say that the CPI is wrong, that doesn't tell you anything. What I'm looking for is is it roughly correct? And you can definitely make methodological adjustments to make it a little bit higher, a little bit lower, but if it's roughly correct to me that's good enough. And I also notice that even in countries that maybe don't have the best governance, maybe say Latin American countries that have high inflation and so forth, even over there the CPI shows higher inflation, right? It's not like you go to Brazil and Argentina and suddenly inflation is magically 2% because it's calculated by the government. So I don't think it's a legitimate concern to say that the US government will suddenly manipulate CPI in a big way. I'm sure they'll do to some extent, but that's probably like a few tenths or something like that. In any case, not giving you any investment advice, but I just point out that if you are someone who's looking at fixed income, we're at levels both in real and nominal that seems to be drawing more demand.
### 美联储政策鹰派转向与精炼油产能的通胀约束 在货币政策预期上，市场与美联储公开表态近期出现了显著的鹰派转向。过去一周美联储理事**克里斯托弗·沃勒**（Christopher Waller）发表讲话，明确否定了 10 月降息的可能性，转而支持在 12 月启动加息，甚至可能在后续季度节奏中追加一次。沃勒的核心逻辑在于：去年美联储实施的 75 个基点**预防性降息**（Insurance Cuts: 在经济尚未陷入衰退时提前释放流动性的防御性政策）事后被证明并非必要，当前劳动力市场依旧坚挺，而再通胀压力正在抬头，因此美联储倾向于收回先前的宽松额度。驱动这一立场改变的支柱包括超预期的 AI 基础设施投资、居高不下的原油价格以及持续的关税摩擦。 然而，通胀的真正硬约束并不在原油开采端，而在下游的工业炼化能力。尽管国际原油价格在每桶 100 美元附近徘徊，终端消费的核心却是柴油和汽油等精炼产品。目前**柴油价格**已飙升至每桶 200 美元的高位，成为供应链成本的最主要瓶颈。精炼产能的紧张一方面由于中东物流地缘政治阻滞，另一方面在于乌克兰对俄罗斯炼油基础设施的持续打击。单纯调降燃油税或寻求海外临时供应无法填补结构性缺口。随着美国中期选举临近，白宫承受着巨大的民意压力，军事冲突导致的全面破坏（MAD 模式）不符合任何一方利益，各方通过外交妥协缓解中东紧张局势的概率正在上升；若地缘风险能达成外交化解，将直接驱动基准利率回落并引爆权益资产的新一轮上行。
Original English Now in addition I also know that we have some interesting new Fed speak and it seems like the consensus now is that there will not be any movement in October and now the market and Fed speak is supporting a December hike. We had Governor Waller come out the past week and as we noted I think last week or two weeks ago we already had Williams and Jefferson say that there's not going to be an October hike and now we have Waller strongly coming out and saying that yeah no October hike, probably a December hike and maybe after that one more. It seems like his thinking is that we cut 75 basis points last year as an insurance cut. Actually we didn't really need that insurance since the labor market was not as weak as we thought. Inflation is creeping higher and so he seems to want to take back those 75 basis points of insurance cuts and so but not consecutively and so I think the base case right now is that we'll just have two more hikes but maybe on a quarterly cadence. Now I found this speech by Waller pretty surprising because just last month he was talking about just give this inflation a chance and so forth and now he's saying he's changed his mind for three reasons. One is of course that the AI buildout better than expected. Of course we have oil prices continuing to go higher which is true and of course tariffs that seems to be persistent, right, between then and now we seem to have more of a trade war with Canada and so forth. And so that's his rationale. I suspect that maybe there's some things happening behind the scenes that affected him last time around because I think of his last speech as more out of character. Now, if you look at oil prices, you can see that we've stabilized at around $100 and there's so much noise here going back and forth between what's happening in the Middle East and what the president is threatening. So, sending more ships over there and so forth. I think the past week the president said that he would not attack Iran before the midterms, obviously reacting to the sudden spike in crude oil prices as news came out that he's sending another ship over to the Middle East. But remember, we don't consume crude oil. We consume refined products like gasoline and diesel. If you look at diesel, it's already at $200 a barrel. Very, very high. So that's actually the real bottleneck right now. Crude oil can go up and down, but again what we consume matters and we don't have enough refining capacity. Part of it is because a lot of refining capacity is in the Middle East and they can't get out. Part of it is of course what's happening in the Ukraine war. Ukraine is attacking a lot of Russian refineries. President's trying to do something about this: lowering gas, lowering diesel taxes. Of course making a deal with Putin to get more diesel out, but it really doesn't seem to going to make a big difference. We really need to have some resolution in the Middle East for rates to come down and if that were to happen of course I continue to think that we would have a huge surge in the stock market. I also am increasingly leaning towards the possibility that something gets worked out before the midterms to save the Senate. I don't believe that we will go back to military action. Not because the US military is not strong, but first off President Trump has had many opportunities to do this, and secondly it's very clear that we are in a mutually assured destruction mode whereas that if Iran gets bombed they will just shoot missiles and destroy all the other oil infrastructure in the Middle East, Saudi Arabia, Qatar, UAE and so forth and it will just make things worse. So I don't think we're going to have any more Middle Eastern hot war conflict like go back to bombing and super clear that there's no appetite for that. So if that's the case then we're just going to have to work out a diplomatic deal and it's better for both sides to do this now than later on.
### AI 资本开支泡沫、营收天花板与电力物理硬约束 在宏观波动之外，AI 投资交易（AI Trade）近期出现了明显的脆弱性迹象。英国《金融时报》（FT）披露关于 **OpenAI** 营收不及预期的报道，在盘中迅速引发了整个 AI 产业链的大幅抛售，暴露了资本市场高度敏感的神经。物理学家、风险投资人**史蒂夫·徐**（Steve Hsu: 理论物理学家，长期关注前沿技术演进）在播客中对该产业链的技术经济学（Unit Economics）进行了系统拆解。目前由**超大规模云厂商**（Hyperscalers: 拥有海量算力与基础设施的顶级云服务商）驱动的 **AI 资本支出**（AI Capex: 用于购置 GPU、存储及建设数据中心的巨额固定资产投资）已攀升至数万亿美元规模，支撑了英伟达、存储芯片等半导体个股的空前繁荣。 然而，要支撑如此庞大的基建投资回报率，软件与大模型企业需要创造出呈几何级数增长的现金流。现实情况是：全行业**代币推理单价**（Per-token Cost: 运行模型生成单个文本单元的边际收费）随着模型迭代迅速跳水，直接削蚀了边际收入；大模型头部玩家虽然前期营收体量可观，但增长曲线已明显放缓，如 **Anthropic** 等企业每年仍面临数百亿美元的巨额净亏损。为使资本支出合理化，整个软件行业的 AI 收入必须在数年内实现十倍级扩张，这在商业化实践中面临巨大鸿沟。更具破坏性的现实约束在于**电网基础设施容量**（Power Grid Capacity: 输电网与发电侧对高负荷数据中心的支持极限）。即便硬件采购到位，美国现有电力建设周期的滞后已导致大量数据中心面临无法通电开工的窘境。虽然 AI 确属划时代的底层技术，但在营收证实与电力供给的双重瓶颈下，当前极度拥挤的资本估值仍面临剧烈的修正压力。
Original English Okay. The second thing I want to talk about is some cracks or at least a little scare in the AI trade. So, as many of you know, I firmly believe the AI is a huge bubble that only tells you that how it's going to end. It doesn't really tell you how high it can go. In fact, it may be the only way out of this is through. And that is to say, we just have some kind of blowoff top. In any case, I came across this very nice presentation by Steve Hsu of Manifold podcast, basically breaking it down. I go for a little bit of it, but I do recommend you guys to take a look at it. Steve Hsu is a very high IQ person, graduated from Caltech at the age of 19, PhD in physics and also actually has some AI startups and stuff like that. So basically he's summarizing the views that many people basically already know. If you're not aware, you can see that the calculations done by some large investment banks are showing you that the AI capex as we all know is enormous and it is a large reason why we have memory stocks and chip stocks doing so well. A lot of money is being spent in the AI buildout. But if you're looking at this from just kind of a return on investment perspective, in order to have return on these trillions of investments, you're going to have to have a lot of revenue by these AI companies. And it could be through the models, through the clouds and so forth, although of course they're intertwined because a lot of the model companies like Anthropic and so forth are big clients to the hyperscalers. Now the point being made by Steve and by many people in the industry is that the amount of capex spent is very large and you need a lot of revenue to be able to justify that. Now if you look at the amount of revenue that the AI companies are getting or just look at the amount of revenue that software companies in the world are getting, it's just never going to make sense, right? You're going to need to have to increase this AI revenue by like 10x in a few years to be able to make the investment make sense and that's just something that doesn't seem to be happening. In addition, what we've seen is that even though there's more and more tokens so people are using more AI, the per token cost is going down so that hurts revenue. And of course as we've discussed in the past, so if you look at Anthropic, they had strong revenues, very large revenues, and historically speaking for a company to go from beginning to where they are right now and generate so much revenue is historic. But we also notice that the revenues don't seem to be growing anymore and they continue to lose money, like a lot, like 40 billion last year I believe. So that is something that suggests that they're not growing revenues fast enough to justify all the AI capex. Now, what was shaking the market the past week was that we had reporting from the FT that suggests that OpenAI's revenues may be lower than expected. Now, there's a lot of debate about this because AI companies, they calculate revenue in different ways. It's not a standardized calculation methodology, so there's dispute about this. But the first knee-jerk reaction from the markets when the news broke was that they sold everything AI related bigly. And that tells me that there's a lot of nervousness. Most sophisticated people in the markets know that there's tremendous amounts of overvaluation, there's some kind of earnings bubble and that things are fragile. However, we also know that it can go higher and higher than anyone expects. So, at the first drop of a hat, they are inclined to sell. Now afterwards of course the market recovered and everyone is thinking that yeah maybe the revenue metrics are not as bad as you thought and you know maybe FT is doing a bad job reporting and so forth. But in any case I think this just emphasizes the fact that there is I think a lot of jitteriness to this and AI is a transformational technology, it will have big impacts, but that does not mean that it will justify investments and that these companies continue to have the valuations that they have. Yet a lot of things happening, and one other thing that many people note when concerns about the trade is that the US physically doesn't have enough electricity to power all the data centers that have been promised. So you could have all these data centers that maybe you buy the chips and so forth, but you just can't operate them because there's not enough electricity. The power buildout is slow and that of course is going to have some constraints on this. So any case that's something to be aware of. Recommend you guys to take a look at that presentation. All right. So that's all I prepared for today. In the coming weeks, again, let's focus more on what happens in the Middle East. And I really hope that we get some kind of resolution because that would be really good for the global economy, really good for yields, and of course, really good for all financial assets. Talk to you guys next week.