CoinW Research Institute

The Federal Reserve will announce the results of the Federal Open Market Committee (FOMC) meeting on September 16, and the Bank of Japan will announce its interest rate decision on September 18. Just before these two meetings, Trump's latest conversation with NVIDIA CEO Jensen Huang at the All-In Summit has brought AI capital expenditure back into the market spotlight. Trump openly opposed slowing AI development due to security concerns, describing data centers as the "oil" of the next 20 to 25 years, while Huang stated that he would not allow the pace of AI development to slow down. This stance starkly contrasts with Anthropic's head Dario Amodei's proposal to slow the enhancement of frontier models, and OpenAI’s Sam Altman and xAI's Elon Musk later showed support for that position, leading the market to reassess whether investments in chips, data centers, and power infrastructure will maintain the previous pace of expansion after the models have turned cautious.

This divergence is important because AI capital expenditure has already entered U.S. inflation and long-term capital pricing. High oil prices continue to increase inflationary pressure in the U.S., while large-scale AI construction raises demand for chips, power, equipment, and long-term financing, with fiscal financing and corporate financing both pushing up long-term capital costs. The current market pricing for a 25 basis point rate hike by the Federal Reserve is about 90%, and the yield on the U.S. 10-year Treasury bond briefly exceeded 5%. The high U.S. dollar yields continued to put pressure on the yen, and Japan is facing higher import costs amid soaring oil prices and a weak yen. A Reuters survey found that 66 out of 68 economists expect the Bank of Japan to raise its policy rate by 25 basis points to 1.25%. If Japan can push up the yen through rate hikes, it can alleviate some imported inflation, but the yen’s appreciation will compress the profitability of carry trades, prompting a reallocation of cross-border funds and subsequently impacting U.S. Treasuries and global risk assets.

Therefore, what the market truly faces is a complex feedback loop of capital costs, exchange rates, and AI investment expectations. In an optimistic scenario, after the Fed raises interest rates, long-term U.S. Treasury yields stabilize at high levels, the Bank of Japan promotes an orderly appreciation of the yen, and the expansion signals released by Trump and Huang continue to support AI infrastructure investments, while the caution on the model end does not translate into a noticeable contraction in capital expenditure. In this case, although U.S. dollar financing remains relatively tight, tech firms, related Japanese exports, and funds engaging in carry trades still have time to adjust gradually. In a pessimistic scenario, long-term U.S. Treasury yields continue to rise, the yen appreciates rapidly, and caution on the model end ultimately further depresses AI capital expenditure, with rising U.S. dollar financing costs, exits from carry trades, and downward revisions of tech profit expectations possibly occurring simultaneously, leading to amplified volatility in stocks, gold, and crypto assets.

Inflation and AI Capital Demand Combine, U.S. Treasuries Under Pressure from Fed Rate Hike

The U.S. August CPI inflation data further strengthened the market's expectations for a rate hike at this FOMC meeting. Overall CPI rose 3.4% year-on-year and 0.4% month-on-month, while core CPI increased 2.4% year-on-year and 0.3% month-on-month, with the core monthly increase exceeding market expectations. Among these figures, gasoline prices rose by 3.9% month-on-month, contributing to more than one-third of the overall monthly CPI increase. As of now, Brent crude oil is reported at about $107.6 per barrel, and WTI crude oil at about $103.3 per barrel. Energy prices have directly raised costs for households and businesses; if high oil prices continue to be transmitted to transportation, logistics, and service prices, the rate at which U.S. inflation declines may further slow. In this context, the Federal Reserve needs to provide more convincing reasons to maintain interest rates at current levels, and the market's pricing for a 25 basis point rate hike has quickly risen to around 90%.

In addition to high oil prices, the ongoing expansion of AI infrastructure investment in the U.S. has also supported demand and capital expenditure. The Fed's latest monetary policy report indicates that corporate fixed investment this year has been significantly driven by AI infrastructure construction, with data center construction simultaneously increasing demand for semiconductors, industrial metals, electricity, and other key inputs. AI is expected to reduce some costs by improving productivity in the long run, but during peak construction phases, the demand for chips, servers, electricity, land, and engineering inputs will form real demand first. If investment demand materializes before productivity improves, price pressures and financing demands may remain strong for some time.

This round of debate quickly expanded from model safety to whether AI capital expenditure will continue to expand. Dario Amodei, head of Anthropic, proposed to slow down the enhancement of frontier models, and OpenAI's Sam Altman and xAI's Elon Musk subsequently expressed their support. However, it is noteworthy that on September 15, Trump publicly opposed slowing AI development due to safety concerns during a call with Huang at the All-In Summit, stating that the U.S. cannot halt the AI industry due to security disputes, describing data centers as the "oil" of the next 20 to 25 years. Huang responded that he would not allow AI development to slow down. The two statements created evident tension, with some heads of frontier model companies emphasizing the pace of training and safety boundaries, while NVIDIA and Trump emphasized the continued expansion of computing power infrastructure.

Looking deeper, this divergence also implies a competition for discourse power over the pace of AI development. Frontier model companies can influence new computing power demand through training plans, model releases, and safety assessments, while NVIDIA controls GPU supply, the CUDA ecosystem, and AI infrastructure platforms, recently collaborating with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion in third-party capital for AI infrastructure through an independent financing platform. Trump's and Huang's public opposition to slowing AI development means that even if some model companies are becoming more cautious in their frontier research and development, computing power investments may still continue to expand through inference, applications, open models, and more enterprise clients. Therefore, assessing whether AI capital expenditure declines should not only look at the speed of model releases but also whether the computing power supply and financing systems led by NVIDIA continue to expand.

Before actual capital expenditure shows a notable decline, U.S. fiscal financing and AI enterprise financing continue to occupy long-term capital simultaneously. The Treasury needs to continue issuing bonds to cover fiscal gaps, while large tech companies and data center projects are increasingly raising capital through bonds, credit, project financing, and specialized computing power financing platforms. The credit risks of Treasuries and corporate bonds are different, but long-term investors such as insurance funds, pensions, and asset management institutions need to allocate capital among different assets. NVIDIA's push for computing power to be regarded as a financeable infrastructure asset means that AI construction can continue to absorb long-term funds even without fully relying on cash from model manufacturers, which will create a time lag between a slowdown in model development and a decrease in capital expenditure. The recent U.S. 10-year Treasury yield breaking above 5% is driven not only by oil prices and rate hike expectations but also by the simultaneous increases in fiscal supply, real interest rates, and corporate financing demands. AI financing is not the sole reason for the rise in long-dated yields, but NVIDIA’s continuous broadening of sources for computing power financing makes the demand for this long-term capital more sustainable.

At the same time, as the long-term U.S. Treasury yield rises to around 5%, the Federal Reserve faces more than just the choice of whether to raise rates. The U.S. August CPI inflation data and high oil prices continue to support strong inflationary constraints, but the long-term financing cost itself has already tightened significantly. If the policy signal is soft, long bond investors may worry that inflation will persist longer and demand higher inflation and term premiums; if expectations for a continuous rapid tightening are further reinforced after a rate hike, future interest rates, corporate financing costs, and long-term Treasury yields may also rise further. The current market tends to expect a 25 basis point rate hike, but the real influence on post-meeting asset prices will come from the meeting's statement, economic forecasts, and how Waller defines subsequent policy conditions. The Federal Reserve needs to maintain credibility in controlling inflation while avoiding exacerbating a rate hike into a new round of long-term financing cost increases. Thus, long-term U.S. Treasuries become a crucial link between the two meetings; if the long-term Treasury yield can stabilize at a high level after the FOMC meeting, the pressure on the long-term U.S. dollar financing cost will ease; if long-dated yields continue to rise, the higher yield advantage of U.S. assets will still attract cross-border funds and continue to put pressure on the yen.

High U.S. Treasuries Increase Pressure on the Yen, Japanese Policies Affect Global Capital

After the U.S. long-term Treasury yields maintain high levels, pressure will further be transmitted to Japan through interest rate differentials and import costs. The yield on Japan's 10-year government bonds has returned to around 3%, and a Reuters survey found that 66 out of 68 economists expect the Bank of Japan to raise its policy rate from 1.00% to 1.25% this week. Meanwhile, high oil prices are increasing Japan's energy import costs, and the high yields of dollar assets are weakening the yen's attractiveness, causing further increases in import prices in yen terms. While Japan's real wages are improving, household consumption remains relatively weak; thus, the Bank of Japan has both reasons to continue raising rates to alleviate imported inflation and a need to control the pace of tightening to avoid putting more pressure on domestic demand. If the Federal Reserve’s rate hike leaves U.S. long-term yields still high, Japan may find it difficult to completely eliminate the pressure on the yen with just a 25 basis point hike.

The U.S. Treasury's recent focus on Japanese policy also reflects considerations for the U.S. Treasury market. Bessenette openly supports Japan adopting a more robust monetary policy, emphasizing the need to stabilize inflation expectations through credible policy and avoid excessive fluctuations in the yen; in previous joint interventions for buying yen, the U.S. also promoted the use of FIMA repo facilities, which allows foreign currency authorities like Japan to obtain U.S. dollars with Treasuries as collateral. The significance of this action is to reduce the necessity of directly selling U.S. Treasuries to raise funds when dollar interventions in the exchange rate are needed.

While raising rates resolves the weak yen issue, it will also lead to changes in capital in the opposite direction. The yen has long been an important funding currency for global carry trades, with investors borrowing yen at lower costs to allocate into dollar bonds, stocks, and other high-yield assets. After Japan raises rates and the yen appreciates, the financing and exchange rate costs of these trades will increase, prompting some leveraged funds to exit overseas assets. At the same time, as the yield on Japan’s 10-year government bonds returns to around 3%, the attractiveness for Japanese insurance companies, pensions, and banks to increase their overseas bond allocations will also decrease after domestic bond returns improve. A moderate appreciation of the yen can alleviate import inflation and reduce the need for foreign exchange interventions; if the yen rises rapidly in a short time, concentrated liquidation of carry trades and repatriation of Japanese funds may amplify fluctuations in U.S. Treasuries, stocks, and crypto assets.

The subsequent changes in AI capital expenditure are an important variable linking the long-term capital demand in the U.S. with Japanese exports. Currently, frontier model companies are being more cautious about their R&D pace, but investments in data centers, chips, and power infrastructure are still advancing, and in the short term, the demand for long-term capital from U.S. enterprises is unlikely to decline significantly, providing continued support for Japan's exports of semiconductor equipment, materials, and precision instruments.

Overall, the core of this round of discussions between Japan and the U.S. revolves around how the pressures from high oil prices, AI capital expenditure, and fiscal financing jointly raise long-term capital costs and how these pressures continue to transmit to the yen and global assets. The Federal Reserve needs to maintain a balance between controlling inflation and avoiding a further increase in long-term U.S. Treasury yields, while the Bank of Japan must alleviate the weak yen and imported inflation while avoiding excessive rapid appreciation of the yen, which could trigger concentrated exits from carry trades. If long-term U.S. Treasury yields stabilize at high levels, the yen appreciates in an orderly fashion, and AI infrastructure investments do not show significant contraction, the market will still have time to gradually digest policy adjustments; if U.S. Treasury yields continue to rise and the yen appreciates rapidly while AI capital expenditure begins to decline, high financing costs, cross-border deleveraging, and downward revisions of tech profit expectations may interact additively, further amplifying the volatility of stocks and crypto assets.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。