Investing FAQ

Through May 29, the portfolio returned 5.2% in Swiss francs on a time-weighted basis. This method removes the effect of deposits and withdrawals. A global portfolio with 60% equities and 40% bonds returned 5.4%. The S&P 500 total return was 9.0% in francs. The portfolio also had a drawdown of almost 4% in late March. A drawdown measures a fall from a previous peak.

Three rotation calls worked through May 29 in Swiss francs. Emerging Markets gained 20.7%, US Small Cap 13.2%, and Japan 13.1%. Europe gained only 4.8%, behind US large-cap at 9.9%. The rate call missed because the Federal Reserve did not cut. Bond holdings were flat to slightly negative. Korean and Taiwanese chipmakers, rather than the expected Chinese technology companies, drove the emerging-market result.

By the Shiller cyclically adjusted price-to-earnings ratio (CAPE), yes. CAPE compares prices with ten years of inflation-adjusted earnings. It rose from 39.8 in December to about 42.7 in late May 2026. The December 1999 dot-com peak was 44.2, and the long-run mean was about 17.3. The ten largest companies represented 39.1% of the index. Nvidia alone represented 8.17%, up from 7.2% in December.

No. Through late May, the Federal Reserve had held rates at every meeting. A February-March oil shock pushed US inflation back towards 4%. My December thesis assumed two or three cuts. By spring, markets expected the European Central Bank to move before the Federal Reserve.

It depends on the interest-rate difference between the two currencies. Hedging removes the effect of exchange-rate moves. With both Swiss and Japanese rates low, hedging the yen cost almost nothing. It also removed a drag that had erased most of Japan's local-currency gain. Hedging the dollar cost roughly 4% because US rates were higher. I kept the US Treasury holding unhedged and retained that interest advantage.

Gold rose from 5% to 6%, funded with new cash. It remained the portfolio's only genuine diversifier. Crypto drifted from 4.5% to 3.5% after falling 19% with equities instead of protecting against the decline. I switched the Japan holding into a Swiss-franc-hedged share class. I also moved two individual stocks into the diversified exchange-traded funds that covered their market segments.

On 18 November 2025, the European Supervisory Authorities published the first official list of Critical ICT Third-Party Providers. The authorities are the European Banking Authority, the European Insurance and Occupational Pensions Authority, and the European Securities and Markets Authority. The Digital Operational Resilience Act (Regulation (EU) 2022/2554) created this category and has applied since 17 January 2025. ICT means information and communication technology. The list contains 19 designations. It includes AWS, Microsoft, Google Cloud, IBM, Bloomberg, London Stock Exchange Group, Tata Consultancy Services, and Orange. Lead Overseers supervise each designated provider directly at EU level under Articles 31–44 of DORA. They can impose a daily fine equal to 1% of average daily global turnover for up to six months. Article 28 requires contractual exit strategies. Article 30 specifies the contract terms for critical functions.

The Clarifying Lawful Overseas Use of Data (CLOUD) Act (18 U.S.C. § 2713, enacted March 2018) compels US-headquartered providers to disclose data. This duty applies "regardless of whether such communication, record, or other information is located within or outside of the United States." Microsoft's transparency report for the second half of 2024 records 5,587 US law-enforcement demands for consumer data. Of those demands, 115 were warrants for content stored outside the US. In late 2024, Microsoft France told the French Senate that it could not guarantee non-transfer of EU-hosted data under a CLOUD Act order. This creates two structural conflicts for an EU bank. Article 32 of the EU Data Act requires providers to resist third-country government access to non-personal data. DORA also requires audit rights that courts can enforce in the host jurisdiction.

The legal position remains unresolved as of mid-2026. EU-incorporated entities operate Amazon Web Services (AWS) European Sovereign Cloud with EU-resident staff and dedicated infrastructure. Its first region, eusc-de-east-1 in Brandenburg, was scheduled to launch in January 2026. AWS plans to invest €7.8B through 2040. A legal opinion commissioned by AWS argues that this structure escapes CLOUD Act reach. A Dutch Ministry of Justice memo from February 2025 raised a different concern. It noted that the ultimate parent is Amazon.com, Inc. The CLOUD Act applies to providers with "possession, custody, or control." Control by the corporate parent could meet that test. Only the first contested warrant against an EU-resident hyperscaler subsidiary can settle the question, and no such case has yet occurred.

Brad Smith announced the clause at the Atlantic Council in Brussels on 30 April 2025. Microsoft included it in a contractual "European Digital Resilience Commitment" with EU national governments and the Commission. Microsoft promises to challenge any government order to suspend its European operations. It also promises to escrow source code in Switzerland if it loses. Microsoft France, however, told the French Senate that it could not guarantee non-transfer under a lawful CLOUD Act order. The International Criminal Court also lost access to its chief prosecutor's Outlook account after Executive Order 14203 sanctioned Karim Khan. Microsoft denies that it actively cut off Khan. The court migrated to openDesk anyway. The court-fight clause is a contractual promise, not a statutory exemption.

Article 32(1) of Regulation (EU) 2023/2854 took effect on 12 September 2025. It requires data-processing providers to take "all adequate technical, organisational and legal measures... to prevent international and third-country governmental access and transfer of non-personal data held in the Union where such transfer or access would create a conflict with Union law." Article 32(2) recognises a third-country order only when it is based on an international agreement in force with the EU or relevant Member State. No EU–US CLOUD Act executive agreement exists. A US warrant for non-personal data hosted in Frankfurt therefore creates a statutory conflict that the provider must resist. Most bank transaction metadata is non-personal under EU law and falls within Article 32.

Bank supervisors in Germany, Austria, and Switzerland apply four layers of rules. DORA (Regulation 2022/2554) applies across the EU. It creates direct supervision of the 19 Critical ICT Third-Party Providers. The European Central Bank (ECB) published its Guide on Outsourcing Cloud Services on 16 July 2025. It turns Single Supervisory Mechanism (SSM) expectations into operating requirements for concentration metrics, exit testing, and enforceable audit rights. The German Federal Financial Supervisory Authority (BaFin) also applies national rules. BaFin's Supervisory Requirements for IT in Financial Institutions (BAIT) align German supervision with DORA. The Minimum Requirements for Risk Management (MaRisk), section AT 9, do the same. The ninth MaRisk amendment dates from June 2024. BaFin's March 2024 cloud guidance adds German enforcement detail. The Swiss Financial Market Supervisory Authority (FINMA) has applied Circular 2018/3 since 2018. It is technology-neutral and allows foreign outsourcing only when inspection rights are enforceable in the host jurisdiction. Across DACH, these frameworks converge on three operational requirements. Banks need a documented and tested exit strategy. They need contractually enforceable audit access in the actual jurisdiction. They also need a measurable concentration metric for ICT third-party providers.

Article 28(8) of DORA requires comprehensive and documented exit plans. Banks must test them sufficiently and review them periodically. In practice, supervisors expect at least one tabletop exercise or partial migration every two years. This standard applies to ICT third-party arrangements that support a critical or important function. Tests must confirm data portability and the availability of alternative providers. They must also show that the planned cutover time is realistic. The ECB Cloud Outsourcing Guide of 16 July 2025 treats exit testing as a continuous duty, so one contractual check is no longer enough.

Partially. By May 2026, most technology and infrastructure calls examined here had landed, received support, or remained on track. They concerned test-time compute, GPQA Diamond, power, gas, the Gulf chip pivot, AMD's $1T compute-market forecast, and continued scaling. Most political prescriptions remained unfulfilled or reversed. There was no voluntary consolidation, Congressional trillions, or coordinated democratic coalition. The US government did not invoke the Defense Production Act (DPA) for the proposed project. Export controls loosened. This verdict remains provisional because some deadlines extend to 2027–28 or the end of the decade.

Artificial general intelligence (AGI) means broad human-level or greater cognitive ability. Aschenbrenner's 2027 window remained open but contested in May 2026. Test-time compute gives a model more processing while it produces an answer. This development supported his framework in a way that he did not emphasise. Pretraining, the initial learning from broad data, had slowed relative to reinforcement-learning (RL) post-training. This shift complicates the additive five-order-of-magnitude (5-OOM) framing. Capability gains continued. The open question is whether models reach his 'drop-in remote worker' threshold by 2027. Such a model could perform the cognitive tasks of a remote employee. His investment firm is positioned around that outcome.

Aschenbrenner combines three normally separate positions. He starts with a libertarian presumption against state action. He also reveres two-hundred-year-old institutions in the tradition of Edmund Burke. Finally, he is empirically optimistic that researchers can align advanced AI. The three positions form a series circuit, where each premise supports the next. He argues that artificial general intelligence (AGI) will decide national security and that existing constitutional institutions must absorb it. This Burkean premise produces a Promethean prescription: peacetime industrial nationalisation.

Aschenbrenner was a member of OpenAI's Superalignment team. He has said publicly that OpenAI dismissed him in spring 2024. He says the dismissal came one to two weeks before Ilya Sutskever and Jan Leike resigned. According to his account, exit discussions covered a security memo that he sent to the board. They also covered his decision not to sign the November 2023 employee letter supporting Sam Altman's reinstatement. His account says that his views on AGI nationalisation were another factor. He has said that he declined a non-disclosure agreement (NDA) with a non-disparagement clause. He says that this decision cost him approximately one million dollars in vested equity. OpenAI has not commented publicly on the specifics. OpenAI dissolved the Superalignment team within five weeks.

Situational Awareness LP is the San Francisco investment firm that Aschenbrenner co-founded in mid-2024. Here, LP identifies the firm's legal structure. In interviews, he has named Patrick Collison, John Collison, Daniel Gross, and Nat Friedman as anchor investors. He says the fund turns his framework into capital positions. A long position gains when an asset's price rises. Behind-the-meter power is generated at or near the customer site. AGI-adjacent infrastructure supports artificial general intelligence without producing the models itself. His described positions include semiconductors, power utilities, behind-the-meter gas, and related infrastructure. The 'big bond short' on real interest rates above 10% had not fired by May 21, 2026. None of the named limited partners has independently confirmed a specific position. Every position description comes from Aschenbrenner's public statements.

Aschenbrenner rejects probability distributions and instead 'tells the modal story'. A modal story is a vivid, dated, and falsifiable narrative about the decade. The method is gradable because readers can test dated claims against events. Its weakness is sensitivity to the underlying process. It works better with empirically lawful processes, including scaling curves plotted on logarithmic axes, capital-expenditure aggregates, and power use by the largest cloud companies. It works worse when elections, executive turnover, and coalition politics determine outcomes.

Menlo's December 2025 State of Generative AI in the Enterprise report puts US enterprise generative AI spending at $37B. Menlo and its research partner surveyed 495 buyers about their 2025 generative AI budgets. The figure covers only US enterprise generative AI. It excludes traditional or predictive machine learning and consumer subscriptions. Worldwide vendor-recognized AI revenue is materially larger because Menlo excludes several categories. These include non-US revenue, AI products that do not use generative models, consulting, and channel margins.