Markets Are Pricing Resilience Before Inflation, Liquidity and AI Returns Are Tested

Rising oil, renewed attention to White House contact with the Federal Reserve, a rapid hedge-fund rescue and a $25 billion AI debt offering all turn on the same unresolved question: how much resilience is actually in the system when inflation, liquidity and expected returns are tested at once.

On the day, escalating Middle East tensions and uncertainty around the Strait of Hormuz pushed crude prices higher, equities lower and Treasury yields up. That matters beyond a single trading session because it creates the sort of inflation pressure that can force the Federal Reserve into an uncomfortable choice—particularly while President Trump’s communications with Fed Chair Kevin Warsh are receiving unusual scrutiny.

Enda Curran reported that Trump had called Warsh on an irregular basis since Warsh took office in May. The reporting did not establish that monetary policy was discussed. Curran said there were general indications of topics, but the concern among Fed watchers was less the mere fact of conversations than the lack of disclosure and the wider backdrop of political pressure on the central bank.

Trump’s preference is not in doubt: he has called for lower interest rates since returning to office and has said the US can sustain them. His public treatment of Warsh, however, has been warmer than his treatment of Jerome Powell. At Warsh’s swearing-in, Trump said he wanted the Fed chair to be “totally independent.” More recently, he praised Warsh while attributing the problem at the Fed to “bad people” on the committee.

Michael McKee said the frequency of the contact was unusual by recent standards, though not without historical precedent. Presidents once dealt more directly with Fed chairs: Richard Nixon met frequently with Arthur Burns and, McKee said, persuaded him to ease monetary policy before the 1972 election; Lyndon Johnson brought Fed officials to his Texas ranch. More recently, presidents have generally maintained greater distance. McKee’s comparison was Powell, who met with Trump once and President Biden once.

That question becomes concrete only if the economic data force a policy choice. Warsh is one of 12 votes, and McKee said at least five officials had indicated after the latest meeting that they would have preferred to raise rates. The chair has not, in that sense, redirected the institution on his own. But a chair seen as close to a president demanding lower rates would face a sharper credibility test if inflation persisted and other officials pressed to tighten.

The immediate catalyst was energy. The market graphics showed WTI crude at roughly $77.30 a barrel and Brent around $82.44, while Treasury yields rose across the curve. The reported backdrop was uncertainty over an Iran-Oman arrangement and local-media reports that Iran would seek to bar US and Israeli ships from the Strait of Hormuz, while requiring compensation from “hostile countries” before passage. Tim Stenovec noted that the status quo before the February 28 war involving the US, Israel and Iran had been an open strait in international waters.

Carol Massar raised the possibility that persistently higher energy prices could push the Fed toward rate increases. Stenovec supplied the central limitation: monetary policy cannot produce more oil or directly reopen a disrupted shipping route. It can restrain demand, but it cannot resolve the underlying supply shock.

Curran’s view was that the next policy narrative would be determined by data rather than by the apparent early cordiality between Trump and Warsh. The July employment report was due the next day, followed by CPI, then PCE, with Jackson Hole and a mid-September Fed meeting ahead. A stable labor market and cooling inflation would reduce the immediate conflict. A renewed inflation problem would make it much harder to separate monetary policy from the political pressure surrounding it.

McKee said the jobs report would matter most in the event of a large negative surprise. If unemployment remained at the expected 4.2%, policymakers could conclude that the economy was not slowing materially. A weak job-creation number combined with rising unemployment, however, would make it harder to argue for rate increases even if inflation were moving lower.

The labor market itself was not simply a question of worker availability. McKee described weak employer demand amid uncertainty over tariffs, costs, inflation and consumer spending. Businesses may be operating lean, he said, but they have been reluctant to hire when tariff policy could raise costs or a widening war could weaken demand. Job openings remained, based on the JOLTS data he cited, yet the market had become “kind of frozen.”

Curran identified inflation as the larger risk: another acceleration after years of price increases could become embedded in expectations. McKee named two related dangers—an escalating war that pushes prices sharply higher and damages consumer spending, and an equity market he regarded as highly valued with money still pouring in. A serious market reversal, he said, could weaken confidence and household willingness to spend.

AI is relevant to that inflation calculation, but not yet as a demonstrated economy-wide offset. Curran said there was still little evidence in aggregate productivity data that AI had produced a broad effect. Productivity may have benefited from small-business creation since the pandemic, alongside other possible factors. The more consequential open question is whether AI diffuses widely enough to lift productivity and, by extension, ease the inflation constraint confronting the Fed.

The rescue of Situational Awareness illustrates a different form of vulnerability: a fund can be right about a long-term asset and still face a near-collapse when leverage, concentrated public positions and margin calls demand cash immediately.

Hema Parmar reported that the fund had returned to investing within days of its crisis. It put about $400 million into an unidentified private company on a Monday, following roughly $100 million invested in late July. The company had previously been backed by Sequoia, Parmar said, but its identity and the transaction’s terms were not known.

The size of the investment is notable because Situational Awareness had just removed the leverage that made it vulnerable. Parmar said the firm used proceeds from the Citadel transaction and portfolio sales to satisfy margin calls, take off all leverage and cover its short positions. What remains is a combination of private holdings and a stock portfolio that was not sold—but the split between public and private assets is undisclosed.

That does not establish a permanent pivot to private investing. Parmar was explicit that it was too early to characterize the fund as leaving long-short public equities or becoming a venture-capital investor. The new $400 million position is private, but the fund may still have a sizable public-equity book, and those holdings may have rallied after the Citadel transaction. Its future allocation and appetite for concentrated risk remain unknown.

The fund’s private positions were crucial precisely because they did not face daily mark-to-market pressure in the same way as public securities. Parmar described the private book, including a substantial Anthropic position, as a “saving grace in some ways.” But illiquidity was not a complete protection. Before the Citadel deal, Situational Awareness had considered selling private stakes—including Anthropic exposure—to raise cash quickly enough to meet margin calls.

1. Late July
   Situational Awareness invests roughly $100 million, according to Parmar.
2. Late July
   The fund seeks capital to meet margin calls and reaches a transaction with Citadel while selling portfolio assets.
3. After the Citadel transaction
   Situational Awareness pays margin calls, removes leverage and covers short positions.
4. Monday
   The fund invests about $400 million in an unidentified company previously backed by Sequoia.

Citadel’s position in the episode was very different. Parmar said Citadel had made less than 50 basis points in July before the transaction. Its monthly return rose to nearly 6% afterward, lifting its year-to-date gain from about 5.5% to closer to 12%. The return reflected the advantage of being able to acquire assets at a discount when another investor needed liquidity quickly.

Situational Awareness had grown rapidly after founder Leopold Aschenbrenner published an AI-focused manifesto titled Situational Awareness. Parmar said the document drew considerable attention and helped create interest in the fund. Its investor base also appeared closer to a venture network than to a traditional institutional hedge-fund client roster: founders, venture-capital participants, employees, wealthy individuals and hedge-fund founders rather than principally large institutions.

Parmar said the fund grew to $45 billion before falling to $10 billion. The surviving private portfolio was a major reason it retained that scale.

The structure of the fund limits how quickly investors can vote with their feet. Parmar said there is a multiyear lockup. Private investments are inherently illiquid, and early investors whose lockups expire sooner may not want to leave if they remain ahead since inception. Newer investors, by contrast, may have entered more recently, absorbed considerable losses and still be locked in for years.

The episode is not simply a caution about investing in AI. Parmar noted that other AI-focused managers had a bruising July without suffering a comparable breakdown. Whale Rock was down 22% for the month but remained up roughly 30% to 34% for the year. Coatue was down 8% in its worst month in more than a year while still positive year to date. A Renaissance equities fund run by computers gained more than 9% during the month, Parmar said.

The distinctive risk was the combination of leverage and concentration. Extreme early gains should not be treated as evidence that those risks have disappeared.

Situational Awareness is again able to deploy capital. That is not the same thing as having answered the questions exposed by the rescue: how large the remaining public book is, how its investors view the volatility, whether the fund will resume a similar risk posture, and whether its private assets will remain a buffer or become another source of illiquidity in the next stress event.

Alphabet’s bond sale offered the strongest evidence in the source that investors remain willing to finance AI spending—provided the borrower has the balance sheet, cash flow and existing infrastructure revenue to support it.

The company sought to raise as much as $25 billion through a 10-part US dollar investment-grade bond offering and drew about $115 billion of demand. That appetite stands in contrast to the unease described around some data-center debt, where two of the last three commercial mortgage bond deals funding data centers had to widen pricing to attract buyers. Investors are not treating all AI-linked debt as interchangeable.

Mandeep Singh said Alphabet needs external financing because it has signaled a significant increase in capital expenditure next year. His $300 billion capex figure was a Bloomberg Intelligence estimate, not Alphabet guidance. In Singh’s assessment, operating cash flow can no longer finance the company’s planned investment by itself, leaving Alphabet to use both equity and debt financing.

The rationale for creditors is not that AI’s ultimate application-layer payoff has been proven. Singh drew a distinction between the parts of the buildout that already show returns and the broader productivity story that remains to be demonstrated.

Cloud hosting, cloud rental, chips, memory and compute are the clearest beneficiaries, he said. Companies want tokens and computing capacity, and that demand has accelerated revenue growth for the large cloud platforms. The infrastructure providers are seeing tangible top-line growth even as a broad return on AI applications remains less visible.

Alphabet also has a structural advantage in that infrastructure race. Singh said it designs its own TPUs, deploys them in its own cloud and monetizes that vertically integrated system at better margins than hyperscalers that must purchase Nvidia chips. That combination of chips, cloud capacity, margins and cash generation gives Alphabet more latitude to borrow repeatedly.

Singh expected the bond offering to be a recurring feature rather than an isolated financing event. He noted that Alphabet had previously issued a 100-year bond and contrasted its position with CoreWeave’s. Both are aggressive AI infrastructure spenders, he said, but CoreWeave lacks Alphabet’s balance sheet and cash flow, making comparable oversubscribed borrowing harder to achieve.

The adoption story is less mature outside the infrastructure suppliers. For businesses using AI, Singh said the early return is primarily cost optimization: supporting IT systems without proportionate headcount growth. The proposition is “more with less,” with productivity rather than a dramatic immediate revenue increase as the initial benefit.

Coding agents are the clearest use case in Singh’s view because software-development productivity is visible. He said that 70% to 80% of new code is being written by AI—a claim he used to explain why companies are trying to extend AI-assisted work into other forms of knowledge work. The broader economic effect, however, remains uncertain. That uncertainty is why Curran’s observation about the lack of clear aggregate productivity evidence matters: corporate examples may be persuasive before they are large enough to change economy-wide data.

Alphabet’s financing strength also does not resolve its competitive issue at the model layer. Singh viewed Google’s decision to concentrate AI leadership in Mountain View as a slight negative because it followed high-profile departures, including Jeff Dean and Noam Shazeer. Losing people with that level of credibility is not favorable, he said, even if management hopes the reorganization will accelerate decisions.

The weak point in the Alphabet thesis, according to Singh, is not infrastructure capacity but model-release cadence. Google has chips, cloud and financing capacity, yet Singh said it had been slower than OpenAI and Anthropic to release frontier models and lacked an equivalent to Claude 3.5 or the latest ChatGPT model. The leadership shift reflects pressure to close that gap.

Alphabet can therefore keep spending because the revenue-producing foundation of the AI buildout is already visible to lenders. What remains unproven is whether the wider economy will generate enough application-layer value to validate the scale of investment now being financed.