In a candid and structurally wide-ranging note published August 24, 2026, Nohshad Shah of Citadel Securities examines three interlocking forces shaping the macro and market landscape: Treasury's debt-management intervention at the long end, the physical constraints still governing AI's buildout, and the accelerating political backlash against data-centre expansion. Read together, the three arguments form a coherent thesis: the hardest problems facing investors today are not analytical but political, and the costs of deferring those problems are being quietly distributed to households. The full report is available here.
A Treasury-Led Operation Twist, Not QE
Shah opens with Secretary Bessent's decision to at least double the cap on 10-to-30-year liquidity-support buybacks from $2 billion to $4 billion per operation through November 4. With seven relevant operations remaining on the published schedule, Shah calculates at least $14 billion of incremental buyback capacity, equivalent to roughly 13% of expected quarterly 20-year and 30-year issuance. The distinction between this and conventional quantitative easing is important, and Shah is precise about it: "This is therefore not QE in the conventional sense...the overall stock of government debt does not decline, and system liquidity need not increase. The cleanest framing is a Treasury-led Operation Twist...broadly liquidity neutral, but nowhere close to duration neutral."
What Treasury is doing, in Shah's framing, is removing long-duration paper from private hands and replacing it with short-dated issuance, reducing the duration the market must absorb and flattening the curve. But the signal may matter as much as the flow. "Whatever the formal language around market liquidity," Shah writes, "the intervention suggests that the Administration is uncomfortable with long-end yields and willing to use debt-management policy to lean against them."
The timing alongside U.S. participation in the yen intervention is, in Shah's view, not coincidental. Japan holds approximately $1.12 trillion in Treasuries, around 12% of reported foreign holdings. A weaker yen paired with higher JGB yields raises the incentive for Japanese investors to repatriate capital or reduce unhedged Treasury exposure. Bessent's willingness to support the Japanese Ministry of Finance, Shah argues, may reflect a desire to limit the risk of forced Treasury sales. The two interventions are connected: one constrains the duration reaching private markets, the other contains a potential source of long-end selling.
The harder verdict, however, arrives plainly: "More broadly, this amounts to financial repression at the margin...policymakers attempting to suppress the market signal rather than resolve the underlying contradiction of procyclical easing in the middle of a generational capex cycle at full employment." If bonds are prevented from clearing at lower prices, Shah argues the pressure simply re-routes. "If the adjustment is constrained in bonds, it is more likely to appear through the exchange rate; a weaker dollar then eases financial conditions and adds to inflation through stronger nominal demand and imported prices." The durable solution, in his words, requires "harder choices on fiscal policy and central banks willing to get ahead of inflation...including, if necessary, by hiking rates."
Energised Compute as the True Scarce Asset
Shah's second section argues that the AI bottleneck remains stubbornly physical. Compute, he is careful to clarify, is not simply GPUs. "It is GPUs plus power, data-centre space, memory, networking, cooling, and the expertise to run them. The real scarce asset is therefore not a chip sitting in a warehouse...it is energised, ready-to-use compute." Supply is growing, but demand is arriving faster, and grid connections measured in years mean that even vast capital expenditure programs cannot close the gap quickly.
For hyperscalers, that scarcity has a counterintuitive implication: "Current cash flow understates the earning power of the installed base." Much of today's capacity was contracted in 2024 and 2025, before the strength of demand was clear. As those contracts roll, infrastructure can reprice higher on a largely fixed cost base. "The spend comes first...the pricing, utilisation, and cashflow follow."
Shah addresses the falling-token-price concern with equal precision. Cheaper intelligence makes more use cases economic, while agents amplify consumption: one human instruction can trigger dozens or hundreds of model calls, so the cost per token can fall even as compute consumed per task rises. The bearish outcome, he specifies, is not cheaper tokens per se but "cheaper tokens without materially higher usage." The bifurcation between frontier models and hyperscalers sharpens the point: "Frontier models may retain the highest-value tasks (planning, reasoning, coding, and orchestration) whilst cheaper or open-weight models handle high-volume execution...and the owners of energised compute capture the economics of both."
Permission as the Next Bottleneck
The third section maps the emerging political economy of AI infrastructure. New York has paused incomplete permit applications for large data centres. Pennsylvania is requiring developers to secure local support and fund the infrastructure they need. Texas, hardly a regulatory laboratory, has ordered a full audit of data-centre projects advancing through the grid-connection process. ERCOT is contending with more than 474 gigawatts of connection requests, with data centres accounting for approximately 90% of them.
Shah frames the structural tension clearly: "The benefits of AI are national, dispersed and often years away, while the costs of building it are local, concentrated, and immediate." The central question is therefore shifting from what AI can do to who pays to make it run. Shah holds a firm philosophical view here. "I remain a firm believer that competitive markets provide the most efficient allocation of scarce resources, but one cannot ignore the political reality. Markets operate within rules set by voters and governments...and those rules are becoming more restrictive." The final observation lands with impact: "The next compute bottleneck may not be silicon, or even electricity...it may be permission."
5 Key Takeaways for Advisors and Investors
1. Treasury's buyback expansion is not stimulus. It is a duration transfer that flattens the curve but does not reduce total government debt or add system liquidity. Portfolios should not interpret it as a QE tailwind.
2. Financial repression is a tax on savers and households. When policymakers suppress yield signals rather than address underlying fiscal imbalances, the adjustment eventually surfaces in the exchange rate and inflation. Currency and inflation protection deserve a place in the conversation.
3. Energised, permitted compute is a more durable moat than chip count. The hyperscalers best positioned to finance dedicated generation, absorb permitting delays, and negotiate community agreements are also best positioned to reprice undercontracted capacity as existing deals roll.
4. Token price deflation is not the right variable to watch. Usage volume and elasticity matter more. If compute consumption per task rises faster than price falls, the AI revenue story remains intact regardless of headline token costs.
5. Political risk is becoming a first-order infrastructure risk. Bipartisan resistance to data-centre expansion means that sites with power, permits, and public consent carry a scarcity premium that will compound over time. This reshapes the competitive landscape in ways that favour incumbents over new entrants.
Footnote:
Shah, Nohshad. "Hard Choices." Citadel Securities, 24 Aug. 2026, https://www.citadelsecurities.com/news-and-insights/macro-thoughts/hard-choices/.
