Long-term US Treasury yields remain elevated primarily due to genuine capital demand, with oil prices and AI representing supply-side and structural challenges. Attempting to suppress long-term yields through higher short-term rates would disproportionately harm the non-AI economy first. The AI industry has evolved into a national investment imperative that the US cannot easily halt, meaning high financing demand and elevated real interest rates are likely to persist.
The Federal Reserve currently faces a situation where interest rates are the only available tool. If it were to make a policy shift with signaling significance, it might later need to re-supply liquidity to the Treasury and dollar system through quantitative tools. Kevin Warsh's hawkish rhetoric across various economic dimensions appears designed to cultivate a tough-guy image, yet the Fed and markets remain trapped in a self-reinforcing feedback loop: markets speculate on the Fed's next move, the Fed adjusts its communication based on market signals, and markets then trade on the Fed's reactions. Eventually, the speeches themselves become the source of volatility.
Policy tools cannot substitute for structural reform. In the current environment, if policies themselves only treat symptoms rather than root causes, the Fed's best course is to speak less to avoid errors, while markets should reduce speculation. At the next FOMC meeting, the Fed will likely maintain the July communication framework: the 2% long-term inflation target remains unchanged, but it will distinguish between short-term price shocks and long-term inflation trends. Markets may once again engage in fresh hawkish-dovish trading around these phrases, just as they did in June and July.
The core issue lies on the "numerator side" — whether AI can rapidly convert its massive capital expenditure into revenue. If AI revenue growth proves sufficiently fast, today's high capital spending, heavy financing, and pressure on the non-AI economy can all be rationalized as resource reallocation during the early stages of a technological revolution. However, if AI revenue fails to materialize, all those so-called "costs" will quickly transform into debt, liquidity, and valuation risks.
Warsh's Speech May Sound Hawkish, But It's Not Guidance
While Warsh's remarks at Jackson Hole leaned hawkish, they contained limited new information. His reiteration of the 2% PCE inflation target seemed more like routine procedure. His core assessment criterion — "returning to target clearly and quickly enough" — remains impossible to quantify, and his pledge of "discipline rather than a decision" reads more as a philosophical stance on policy methodology than actionable guidance.
More importantly, his hawkish framework implies that not only is tightening needed now, but even last year's easing was entirely unnecessary. Inflation trends stopped improving from early last year, corporate profit margins continued strengthening, and employment concerns could be attributed to supply-side changes. Before last year's rate cuts, financial conditions were actually looser than they are today.
Judging by the "trends" Warsh references — whether measured through simple counts or weighted perspectives — US inflation improvement peaked in the first quarter of 2025 and has subsequently deteriorated. Under such a scale, last year's 75 basis points of rate cuts should never have happened, which naturally leaves no room for maneuver now. From an "excess inflation" perspective, the excess component (three-month annualized growth versus prior averages) stems from financial services and insurance, while other core services (airfares, healthcare, and communications) and certain core goods also contribute.
To suppress excess inflation, any of the following would be needed: a stock market decline (which would temper compensatory rate increases in insurance), cooling oil prices, administrative intervention in US healthcare pricing, or revisiting tariffs to re-embrace global disinflationary forces. None of these are simple matters, nor can they be resolved by interest rate tools alone. Warsh's invocation of these dimensions naturally casts him as a hawk, but that does not necessarily reflect his future actions.
The Mirror-of-Mirrors Between the Fed and Markets
Another issue is that markets don't truly treat Warsh as a "new chairman." He has repeatedly told markets not to interpret his speeches as forward guidance, yet market participants continue operating on institutional memory. The "mirror of mirrors" dilemma Warsh describes is precisely this trap: markets guess at the Fed's intentions, the Fed treats market pricing as an economic signal, and markets then trade on the Fed's response to market behavior. When both sides treat each other as information sources, prices ultimately reflect mutual speculation rather than economic fundamentals.
The more frequent the speeches, the more likely noise gets mistaken for signal. Years of forward guidance created a market dependency: whatever the central bank says, asset prices react in advance. This worked reasonably well in a relatively stable economic environment. But the current environment shifts too rapidly — wars, supply chains, AI investment, and the international order can all alter inflation and growth paths within months, or even weeks.
In today's climate, policy tools cannot substitute for structural reform, and suppressing industrial cycles cannot fix capital gaps. If the policies themselves only treat symptoms, the Fed should speak less to avoid mistakes, and markets should speculate less. Given the stated intention to reduce forward guidance, public rhetoric can no longer be directly equated with future actions. What a chairman thinks privately, what he says publicly, and what he ultimately does are three entirely different things. When actually making policy, he must contend with inflation, employment, stock market performance, midterm elections, liquidity conditions, the non-AI economy, Japanese FX intervention, Treasury markets, fiscal constraints, and political pressure from the executive branch.
The "Rate Hikes Save Long-End Yields" Logic Is Difficult to Sustain
A common market view holds that if the Fed hikes again, markets would believe it is more resolute in controlling inflation, causing inflation expectations and risk premiums to fall, which would ultimately push long-term Treasury yields lower. This logic holds during cycles when inflation expectations are clearly unanchored, but it does not accurately describe today's Treasury market structure.
Currently, the 10-year nominal Treasury yield stands at approximately 4.75%, the 10-year real yield is around 2.44%, and the 10-year breakeven inflation rate is about 2.3%. For comparison, when Powell delivered his "painful speech" on August 26, 2022, the 10-year nominal yield was roughly 3.03%, breakeven inflation was about 2.57%, and the real yield was only about 0.5%. Compared to then, today's nominal yield is 170 basis points higher, yet real yields have surged by approximately 200 basis points while breakeven inflation has not risen. The primary driver of elevated long-end yields is real rates, so simply hiking to "prove central bank credibility" cannot solve the problem.
On a daily basis, the long end has not truly been trading inflation risk premium — its upward movement has been extremely limited. Instead, it is mostly trading changes in real term premiums. If Warsh continues projecting a hawkish image, markets may deduce a higher policy rate trajectory; under this offsetting dynamic, long-end yields are more likely to be pushed modestly higher — which is exactly what markets ultimately traded after his speech.
A more hawkish Fed can certainly continue suppressing inflation expectations, but with expectations already broadly anchored, the cost would be further elevating short-end real financing pressures. Oil remains a supply-side problem that monetary policy cannot directly address. A 25-basis-point hike cannot restore shipping through key waterways or enhance missile inventories and military capabilities. What truly determines whether oil prices fall is whether geopolitical conflicts de-escalate and whether the US can re-establish control over critical sea lanes — not monetary policy.
If Rates Rise, the "Non-AI" Economy Gets Hit First, Not AI
US AI-chain technology companies' share of non-financial corporate bond issuance (maturities of 1 year or longer) has climbed from 17.9% in 2023 to 37.9% year-to-date through 2026. In the first eight months of 2026, AI tech companies have already issued $300 billion in bonds — more than the combined total of the previous three years. High rates suppress AI last, not first; they hit real estate, manufacturing, consumption, and other cyclical, rate-sensitive non-AI sectors hardest.
If the Fed hikes a little, AI companies can still finance easily, but the non-AI economy weakens further. If it keeps hiking until even AI shows cracks, the traditional economy would likely already be in severe recession. Even what Warsh considers the relatively robust US private domestic final purchases (PDFP) shows latest growth slightly weaker than prior averages. In terms of component contributions, the excess contribution from equipment investment is nearly offset by weakness in housing investment.
Moreover, because AI investment has a high import content, Fed estimates over the most recent five quarters show AI imports offset approximately one-third of AI investment contributions. In Q1 2026, AI investment contributed 1.18 percentage points, but after deducting net imports, this fell to 0.73 percentage points. Critically, regardless of what the Fed says or does, one trend will not change: AI investment and financing will continue to grow.
AI has evolved from a purely commercial investment into a national narrative intertwined with the dollar system, sovereign debt, national competitiveness, and global influence. AI scientists may be borderless, but AI models clearly are not. The US will not voluntarily halt AI investment simply because short-term financing costs are high or the non-AI economy is under pressure. On the contrary, monetary policy, fiscal policy, diplomacy, and even military strategy will increasingly revolve around AI competition. These constraints that transcend the economic sphere are the roots of the Fed's visible sense of helplessness.
High real rates are not a short-cycle phenomenon. The greater AI's financing needs, the more global capital flows preferentially toward high-return technology sectors, forcing other economic sectors and governments to pay higher rates to secure funding. These are all "costs" of the AI era. If AI ultimately delivers high revenue and productivity gains, these costs can be absorbed by future growth. If AI fails, today's high rates, weak non-AI economy, and fiscal pressures will truly transform from "costs" into "risks," and long-term Treasury yields could go even higher.
More Fed Communication Helps Little — AI Revenue Remains the True Variable
Warsh has long supported shrinking the Fed's balance sheet, but in today's environment, that established label actually constrains policy space further. The Treasury Department can temporarily relieve pressure through the Treasury General Account, adjusting debt maturity structures, and Treasury buybacks. The Fed can provide some cooperation through increased short-term bill operations, but matters involving true long-term balance sheet expansion cannot be quickly reversed in the short term.
The problem is that the urgency for such a reversal is rising. If AI financing continues growing and fiscal bond issuance cannot stop, markets will eventually need more base money and liquidity support. At that point, the Fed may be forced to revisit whether to pause or even reverse quantitative tightening — which would sharply conflict with Warsh's past policy identity. Furthermore, the kind of fragility that preceded the Silicon Valley Bank crisis is beginning to re-accumulate.
US banks still carry approximately $320 billion in unrealized losses. While this has declined as a share of total assets, if high rates persist long enough (or rise faster), what are currently static paper losses could eventually morph into liquidity problems. Combined with overseas economies facing greater dollar demand (through imports or FX intervention) that may sell Treasuries to obtain liquidity, these forces could break the current equilibrium.
At the next FOMC meeting, Warsh will most likely maintain the July framework: the 2% long-term inflation target stays intact, but a distinction is drawn between short-term price shocks and long-term inflation trends. He won't fixate on a single PCE or core PCE reading but will observe a basket of indicators. If the long-term inflation trend continues downward, there's no need to obsess over decimals. Simultaneously, fiscal conditions, liquidity, employment, and economic fundamentals must be weighed holistically. Markets may still engage in fresh rounds of hawkish-dovish trading around these phrases, as they did in June and July. But the true core does not lie there — it lies on the "numerator side," namely whether AI can monetize quickly enough. If AI revenue growth is sufficiently fast, today's high capital expenditure, heavy financing, and non-AI economic pressure can all be explained as resource reallocation in the early stages of a technological revolution. If AI revenue fails to deliver, all so-called "costs" will rapidly convert into debt, liquidity, and valuation risks.
Risk warnings: macroeconomic data is reported at low frequency, and estimates for cross-border industry data may lag actual conditions. US industrial policy is significantly influenced by partisan dynamics, and changes in the ruling party could trigger policy reversals. AI technology development and industrial investment carry substantial uncertainty.