The Long Game

The CSIS post-summit assessment published on 20 May reached a pointed conclusion: the Beijing talks "revealed how little progress has been made on the most consequential dimensions of US-China competition: AI, cyber operations, export controls, and digital sovereignty."¹ The trade headlines dominated; the structural contest did not resolve. Huawei's Tau Scaling Law announcement, published five days after Trump departed Beijing, was not a coincidence of timing. Beijing has framed semiconductor self-sufficiency as a national strategic objective, and the Tau announcement lands as a signal that the architectural path to that objective is now publicly declared.

The Iran ceasefire remains formally in place. Tehran's posture since the April 8 agreement has been one of sustained pressure through non-kinetic means: the cable sovereignty doctrine, continued Hormuz signalling, and what CNN described as an escalating effort to demonstrate that Iran possesses powerful levers beyond military force.² Energy pricing reflects this: Brent crude remains elevated from pre-conflict levels, and European gas prices have not normalised. Digital infrastructure underwriting models priced before February 28 still carry energy assumptions that do not reflect the current environment.

On 26 May, Huawei unveiled the Tau (τ) Scaling Law, a proposed architectural framework for achieving transistor density equivalent to a 1.4-nanometre chip process by 2031, without access to ASML's extreme ultraviolet lithography equipment.³ The South China Morning Post reported the announcement as "another DeepSeek moment" among analysts, noting that Nvidia is "likely the most concerned" of the major chip vendors.⁶ If the Tau Law produces results at commercial scale, it would represent a partial bypass of the ASML chokepoint that has underpinned the export control rationale since 2022.

The immediate caveat is distance between claim and delivery. Natixis analysts noted the law "still needs to be tested in practice."⁶ China's current best domestically available AI processor, the Huawei Ascend 950, trails Nvidia's Blackwell-class hardware by a substantial margin on compute performance, though the magnitude varies considerably by benchmark methodology: single-chip comparisons, cluster-level efficiency metrics, and pre-commercial-release estimates produce meaningfully different figures across analyst sources. The directional gap is not in dispute; its size is. The 1.4nm-equivalent target is five years out on Huawei's own stated timeline, and advanced chip production at scale involves manufacturing discipline well beyond architectural law. The announcement is a signal, not a delivery.

The investor implication worth flagging is about hold periods rather than current positions. AI data center assets valued on a 5-7 year horizon implicitly carry an assumption about the durability of US chip leadership. The Tau announcement does not change the near-term picture; it does change what a responsible scenario analysis looks like for positions with that duration.

The Iranian Armed Forces operational command's stated intent to charge transit fees for undersea cables passing through the Strait of Hormuz is being developed into a legal-strategic posture distinct from the kinetic threat. CNN reported on 17 May that Iran is "increasingly signalling that it has powerful tools at its disposal beyond military force" and that the cable framing is part of a deliberate effort to demonstrate non-kinetic leverage, particularly as fears of conflict resumption have grown since Trump's return from Beijing.²

The practical implications are layered. Seventeen subsea systems pass through or near the Strait of Hormuz. The FALCON, Gulf Bridge International, and SEA-ME-WE 5 systems carry the bulk of India-to-Europe and intra-Gulf traffic. An Iranian sovereignty claim would, if pursued legally, create contested jurisdiction over cable operations, maintenance access, and repair windows, entirely independently of any physical threat to the cable itself. Alcatel Submarine Networks' force majeure on 2Africa Pearls remains in effect.⁷

The risk assessment for Gulf cable corridor assets now needs two parallel tracks: the established kinetic disruption scenario, and a legal/sovereignty scenario in which Iranian regulatory demands become a cost and compliance variable for cable operators. The two scenarios are independent; either can materialise without the other.

The US Department of Energy published its one-year review of the Trump nuclear executive orders on 23 May, confirming that three of the eleven projects selected for the Reactor Pilot Programme have secured a Final Documented Safety Analysis as of May 2026.⁴ The programme, established under EO 14301, directs DOE to authorise at least three advanced nuclear test reactors to achieve criticality by July 4, 2026, using DOE authority under the Atomic Energy Act rather than Nuclear Regulatory Commission licensing. DOE Secretary Wright has acknowledged that "one or two" designs are most likely to meet the July 4 date.

Aalo Atomics and Antares Nuclear are the most advanced publicly documented participants. Aalo Atomics is constructing its Aalo-X reactor at Idaho National Laboratory with a target of zero-power criticality by July 4, and has described plans to site an experimental data center adjacent to the reactor as a first-of-kind demonstration of DOE-authorized nuclear power for AI infrastructure.⁹

Two features of the programme are material for PE investors in digital infrastructure. First, the EO explicitly designates AI data centers as critical defense facilities, directing DOE to use all available legal authority to site, approve, and authorize advanced reactors to power them. Second, the DOE authorization pathway bypasses NRC licensing, compressing timelines in ways that change the commercial calculus for captive nuclear power as an investment thesis. The gap between a test reactor achieving criticality in July and a commercially operational captive power asset for a specific data center campus remains large. The point is that the institutional and regulatory architecture for closing that gap is now being built in real time.

CNBC's 24 May analysis of the Gulf AI buildout under war conditions confirmed what several signals had been pointing toward for months: the Iran conflict has produced a visible split between sovereign-backed platforms and commercially operated data center investors in the region.¹⁵ Before the war began in February, the UAE, Saudi Arabia, and Qatar had positioned themselves at the centre of the US AI expansion, leveraging cheap energy, sovereign capital, and strategic geography to attract hyperscaler commitments running into the hundreds of billions of dollars.

The conflict changed the proposition. Two AWS data centers in the UAE were struck by drones early in the war; a third in Bahrain was damaged by a nearby blast. Iran's IRGC struck Oracle's Dubai facility on 3 April, claiming the sites were supporting "the enemy's military and intelligence activities." The framing matters: commercial cloud hubs are being described in Iranian state media as legitimate wartime targets.¹⁵ Pure Data Center Group CEO Gary Wojtaszek told CNBC the company had temporarily paused investment decisions in the Middle East. BCLP partner Mark Richards described decisions as "taking longer because of the nature of the risks."

The sovereign platforms are not pausing. G42 told CNBC its "direction remains unchanged" and its "conviction has only deepened." Amazon's CEO described the company's long-term excitement about Gulf investment as "just as strong as it's ever been." MEI's analysis published 24 May notes that Gulf missile and drone defence systems intercepted approximately 94-95% of Iranian strikes through early May, and concludes that physical defence of AI data centers is "eminently feasible at present," though the analysis notes the existing defensive architecture was not designed with gigawatt-scale AI clusters as a primary protected asset class.¹⁶

The CFIUS read-through is direct. Fund structures with Gulf sovereign co-investors in US AI-adjacent assets now carry a compounded question: the same sovereign entities pressing ahead with Gulf AI investment despite active conflict are simultaneously deploying capital into US infrastructure. A CFIUS reviewer looking at MGX, Mubadala, or PIF co-investment in a US AI data center platform this week is reviewing those relationships in a materially different geopolitical context than existed in Q4 2025. The sovereign/commercial split does not resolve that question; it intensifies it.

• 01
  For AI data center assets with holds exceeding five years: has the underwriting model been stress-tested against a scenario in which Huawei's Tau Scaling Law delivers commercially viable near-parity chip performance by 2030-31, and what is the revenue impact if Chinese customers migrate from US-chip-dependent infrastructure to domestically supplied equivalents?

  Huawei Tau Scaling Law announcement, SCMP, 27 May 2026
• 02
  For Gulf and Red Sea cable corridor assets: has a legal assessment been obtained on the standing of Iran's claimed transit fee jurisdiction under UNCLOS and existing cable landing agreements? Is there a force majeure position that covers not just physical disruption but legally contested access?

  Iran cable sovereignty doctrine, CNN, 17 May 2026; Bearingdrift analysis, 27 May 2026
• 03
  For fund structures with Gulf sovereign co-investors (MGX, Mubadala, PIF) in US AI-adjacent positions: has the sovereign/commercial investment posture split documented this week been factored into the CFIUS assessment? Specifically, does the co-investor's continued commitment to Gulf AI buildout under active conflict conditions change the risk narrative for the US-side investment structure?

  CNBC Gulf AI stress test, 24 May 2026; MEI Iran war Gulf AI analysis, 24 May 2026

• 04
  For Gulf-region data center assets: has the business continuity and connectivity plan been updated to address legally contested access to cable infrastructure as a distinct scenario from physical cable disruption? What is the RPO and RTO under each scenario?

  Iran sovereignty doctrine escalation; cable sovereignty legal framework
• 05
  For all data center assets underwritten before February 28, 2026: has the energy cost model been updated for current Brent and European gas prices? What is the IRR sensitivity across the hold period at current versus pre-conflict energy pricing?

  Iran war energy shock; no price normalisation post-ceasefire
• 06
  For assets pursuing captive power solutions: has the DOE reactor pilot programme pathway, under which AI data centers are designated critical defense facilities, been factored into permitting strategy? What is the current siting status and has DOE's expedited review authority been engaged?

  DOE nuclear EO one-year review, 23 May 2026; AI data center designation as critical defense facilities

• 07
  Does the asset or its customers have material revenue dependency on supply chain relationships that the BIS 50% Rule, effective November 10, 2026, would affect? Has external export compliance counsel been engaged to complete the ownership-chain mapping required?

  BIS 50% Rule, November 10, 2026 enforcement date
• 08
  For US data center assets in Wisconsin, North Carolina, Texas, Virginia, or Ohio: has the full-cost-of-grid-service tariff scenario been modelled, and has the IRR impact been stress-tested against the Wisconsin We Energies precedent being replicated in the asset's operating state?

  US state power tariff reform; Wisconsin and North Carolina precedent

• 09
  For positions using the I Squared / Cogent transaction as a market pricing reference ($4.2M/MW implied entry before conversion capex): has the conversion capex to liquid-cooled high-density specifications been independently estimated, and at what utilisation level does the repositioned asset generate returns competitive with lower-entry alternatives elsewhere in the pipeline?

  I Squared Capital / Cogent Fiber, 25 May 2026; ACG Insights PE Weekly, 28 May 2026
• 10
  For power-for-digital positions that include advanced nuclear in the investment thesis: what does the path from DOE-authorized test reactor criticality (July 4, 2026 target) to commercial-scale captive power offtake actually look like in timeline and capex, and which of the 11 DOE pilot programme participants has the most credible near-term commercial data center offtake plan?

  DOE Reactor Pilot Programme; Aalo Atomics data center adjacency plans; DOE one-year review, 23 May 2026

• For fund structures with Gulf sovereign co-investors in US AI-adjacent assets, initiate a CFIUS posture review against the current conflict environment. The combination of the sovereign/commercial investment split documented this week, the drone strikes on commercial hyperscaler facilities, and the ongoing ceasefire uncertainty constitutes a materially different context from Q4 2025. Proactive review is more defensible than reactive response if CFIUS attention increases.
• Commission legal assessment on Iran's cable sovereignty doctrine for any portfolio asset with Gulf or Red Sea corridor cable dependency. The risk framework for these assets needs a legal/sovereign analysis track alongside the existing kinetic track. The two scenarios are independent: each can materialise without the other, and the governance requirements differ.
• Confirm BIS 50% Rule compliance programmes are engaged and underway. November 10 is five months away. The complexity of ownership-chain mapping for the expanded restricted party definition means external counsel needs to be working now, not in Q3.

• For captive power strategies, engage with the DOE reactor pilot framework as an active siting and permitting pathway. The formal designation of AI data centers as critical defense facilities under EO 14301 creates an expedited review authority that did not exist two years ago. Value creation theses that include captive nuclear should be testing this pathway, not waiting for NRC commercial licensing timelines.
• Add a Huawei Tau Law scenario to underwriting models for AI data center assets with holds beyond 2030. The specific question: what is the revenue impact if Chinese customers migrate to domestically supplied AI compute at near-parity performance between 2029 and 2031? This does not change near-term valuations. It does change what a responsible exit narrative needs to address at deal committee level.

• Develop a framework for re-rating AI data center assets whose investment thesis implicitly relies on a durable US-China chip advantage. The Huawei Tau Law is not a 2026 event; it is a 2031 scenario that needs to be incorporated into the planning of positions being underwritten today for 5-7 year holds. The framework does not need to assume Huawei succeeds. It needs to account for the possibility.
• Monitor the Gulf sovereign/commercial investment bifurcation as a leading indicator for CFIUS and exit market conditions. If the sovereign platforms (MGX, PIF/Humain, G42) maintain their AI buildout commitment through the conflict while commercial operators remain paused, the divergence will narrow exit optionality for assets requiring Gulf sovereign buyers and complicate CFIUS posture for assets with Gulf sovereign co-investors. The I Squared / Cogent transaction ($4.2M/MW entry, 9 US markets) provides a current reference point for domestic AI infrastructure pricing absent Gulf capital.