The Beijing Test

The week is defined by two simultaneous geopolitical events. Trump meets Xi in Beijing on 14–15 May, the first US presidential visit to China in nearly a decade, while ceasefire negotiations with Iran continue under acute strain. On 12 May, Trump described the ceasefire framework as on "massive life support" and flagged the option of resuming combat operations.¹ The Strait of Hormuz remains constrained: mines laid during the conflict are still present (reports indicate Iran has lost track of some of them) and Operation Project Freedom (the US Navy escort mission launched 4 May) was suspended on 6 May after Trump cited "great progress" in diplomatic discussions.² No formal reopening of the strait has been declared.

Brent crude is running at approximately $100 per barrel, sustained well above pre-conflict levels, and the energy price transmission into data center operating costs is now compounded by the transformer supply chain structural gap covered in §7. The AI chip policy backdrop ahead of the summit remains shaped by the April 30 OSTP distillation warning noted in Issue 008: the range of outcomes from Beijing widens the policy uncertainty around H200 export controls rather than narrowing it.

Trump travels to Beijing for a summit with Xi Jinping on 14–15 May, accompanied by a delegation that includes Elon Musk, Apple CEO Tim Cook, BlackRock CEO Larry Fink, and Blackstone CEO Stephen Schwarzman.³ The guest list signals that AI infrastructure financing and technology-sector investment are at the centre of the US delegation's commercial interests. There are conflicting reports as to whether Nvidia CEO Jensen Huang is attending: Euronews cited an Nvidia spokesperson confirming his attendance at the invitation of the President; The Hill, citing a separate source, described his absence as "notable."⁵⁶

The substantive technology agenda, as reported ahead of the summit, includes AI chip export policy, the creation of a bilateral AI safety communications channel, rare earth supply chains, and the Hormuz situation. A chip-specific agreement appears unlikely: Secretary of State Rubio has described AI chips as a "red line." China itself has been discouraging adoption of H200 chips by its own AI labs, directing demand toward domestic hardware (Huawei Ascend), limiting the commercial case for a framework reversal even from the Chinese side.⁴ Chatham House notes that both Washington and Beijing have a shared interest in reopening the Strait of Hormuz; if Beijing can facilitate Iranian concessions, this summit becomes a relevant variable for cable route risk as well as chip policy.¹²

The investor implication: deal teams with AI data center positions dependent on H200 procurement under the current BIS framework should treat this summit as a policy risk event rather than a positive catalyst. A bilateral AI safety agreement would be a positive signal for the sector broadly; a harder chip stance (if the distillation warning narrative prevails) would compress the exit buyer universe for positions with Chinese customer revenue chains.

The Hormuz ceasefire framework, declared 8 April following the February 28 US-Israel strikes against Iran, remains under severe strain. Trump described it on 12 May as on "massive life support," having grown impatient with Iran's continued closure of the strait and what his administration characterises as division in Iranian leadership preventing substantive nuclear concessions.¹ Both parties have fired on each other since the ceasefire declaration. Iran has reportedly lost track of some of the mines it laid in the strait, complicating mine-clearing operations and raising the baseline risk even under a technical ceasefire.

Operation Project Freedom, the US Navy mine-clearing and commercial escort mission launched 4 May, was suspended on 6 May following what Trump described as "great progress" in diplomatic discussions; Iran's military warned the mission constituted a ceasefire violation.² The structural risk picture from Issue 008 is unchanged: 17 cables through or near the strait, 2Africa Pearls force majeure in effect, 63 cable repair ships globally with two to four in the Middle East, stationary repair not viable in a mine-contaminated environment. The directional signal this week is marginally less negative: active negotiations, US Navy presence, and the Beijing summit's Hormuz sub-agenda all point toward eventual resolution. Active Watch is maintained.

The context: on 20 April 2026, the White House issued Presidential Determination 2026-10 under Section 303 of the Defense Production Act, formally finding that US grid infrastructure (including transformers, substations, high-voltage circuit breakers, and electrical core steel) is "essential to national defense" and that "domestic industry cannot reasonably be expected to provide these capabilities in a timely manner."⁸ The determination activated DOE financial tools (purchases, purchase commitments, loans, and loan guarantees) to support domestic production capacity expansion. What it did not do was quantify the problem it was responding to.

This week, the Coalition for a Prosperous America published that quantification - and the numbers are the investor-relevant development.⁷ Before 2020, large power transformers carried 24-30 month delivery lead times. That has stretched to as long as five years. US transformer imports have doubled from $16.1 billion to $35.4 billion between 2020 and 2025. More than 8,000 high-power transformers were imported from China through October 2025, up from fewer than 1,500 in all of 2022. A group of US utility executives visiting a Chinese transformer factory in January 2026 found approximately half the production floor destined for the US market. Crusoe Energy, building the OpenAI 1.2 GW campus in Abilene, Texas, has resorted to refurbishing transformers from shuttered power plants as a stopgap. Citrini Research notes that GE Vernova's Q1 2026 electrification backlog net additions were nearly as large as all annual additions from 2022 to 2025 combined; the backlog is accelerating, not abating.⁹

The CPA analysis also puts a number on the policy response gap: approximately $323 million remains available in the DPA Fund for FY2026, modest against a supply chain importing $35 billion annually and carrying five-year lead times.¹⁰ The marginal supplier of large power transformers to the US grid is currently Korean; allied-market diversification is the medium-term strategy, not a near-term solution. The DPA invocation told the market a problem existed. This week's analysis tells it how large the problem is. For AI data center development assets, transformer and switchgear procurement status is now a critical path item that can be quantified - not merely flagged.

Two transactions this week. Nscale secured $790 million in additional financing for its Narvik, Norway AI data center on 11 May, in a five-bank consortium comprising ABN AMRO, DNB, Eksfin (Export Finance Norway), Nordea, and SEB, the largest AI infrastructure investment in Norway.¹¹ This follows Nscale's $2 billion Series C in March 2026 and a $1.4 billion delayed-draw term loan in February. Sovereign export credit participation, Norway's hydroelectric grid, and a location entirely outside Gulf and Red Sea conflict risk zones make this a data point worth tracking for European institutional investors seeking alternatives to US-concentrated AI infrastructure.

Nvidia announced on 9 May a right to invest up to $2.1 billion in data center operator IREN alongside a 5 GW DSX infrastructure commitment, and a separate $3.2 billion investment in Corning for three US optical fiber production facilities. Both are consistent with Nvidia's strategy (now totalling more than $40 billion in equity commitments, including $30 billion in OpenAI) of taking positions across the AI infrastructure stack as a commercial flywheel.^12b Mizuho analyst Jordan Klein's note on the neocloud deals captures the underwriting question precisely: "it smells like you are pre-funding the purchase of your own GPUs." When a chip vendor holds meaningful equity stakes in its own anchor customers, the demand signal from those customer commitments is no longer fully independent of the vendor's balance sheet.

For context on where the GPU-backed debt market is heading: CoreWeave's $8.5 billion investment-grade GPU-backed ABS (the first of its kind, closed in April) introduces a structural question that will recur across the sector: how does a GPU lifecycle of approximately seven years interact with a data centre lifecycle of 20-plus years in long-dated securitisation structures? JPMorgan projects $30–40 billion in annual data center ABS/CMBS issuance in 2026 and 2027; whether that market can absorb mega-AI structures at scale is being tested in real time.

• 01
  For AI data center assets dependent on H200 procurement under the current BIS case-by-case framework: what is the revenue impact of a reversion to "presumption of denial" following the Trump–Xi summit? Does the offtake agreement contain a material adverse change or force majeure clause triggered by a US export control policy change?

  Trump–Xi summit AI chip agenda — Fox Business, 12 May 2026; The Hill, 13 May 2026
• 02
  For any asset with Gulf or Red Sea cable corridor dependency: does the current SLA include a force majeure carve-out covering an active ceasefire with contested mine clearance in the Strait of Hormuz? Has the insurer confirmed their current position following Operation Project Freedom's suspension?

  Iran ceasefire "massive life support" — CNN, 12 May 2026; 2026 Strait of Hormuz crisis — Wikipedia, updated 13 May 2026
• 03
  For US data center development assets: has transformer procurement been confirmed with a purchase order and specific delivery commitment? What is the construction schedule impact of a 36–60 month delay in transformer delivery, and has the IRR been re-run under that scenario?

  Coalition for a Prosperous America — transformer supply chain, c. 6 May 2026; White House DPA determination, 20 April 2026

• 04
  For US data center development assets: does the power solution include on-site or captive generation that bypasses the grid interconnection and transformer bottleneck entirely? If the asset is grid-dependent, has the local utility confirmed transformer availability and energisation timeline against the construction schedule?

  DPA determination 2026-10, 20 April 2026; Utility Dive — DPA transformer impact, 23 April 2026
• 05
  For assets with Nvidia as a significant supply-chain or equity partner: what proportion of contracted capacity depends on commitments from Nvidia-equity-backed customer entities versus independently contracted end-customer demand? Has the deal team mapped the demand signal to contracts without Nvidia balance sheet involvement?

  CNBC — Nvidia $40B+ equity bets, 9 May 2026

• 06
  For US data center development assets: does Presidential Determination 2026-10 (DPA Section 303, 20 April 2026) create federal procurement preference obligations affecting the asset's transformer or switchgear supply chain? Has external counsel assessed whether existing procurement contracts with Chinese manufacturers are affected by any activated DPA authority?

  White House Presidential Determination 2026-10, 20 April 2026; Burr & Forman DPA analysis, 8 May 2026

• 07
  For Nordic or European data center assets seeking institutional debt: does the Nscale $790M, five-bank consortium (ABN AMRO, DNB, Eksfin, Nordea, SEB) set applicable pricing benchmarks for this asset? What PUE target and green certification criterion enabled Norwegian sovereign export credit (Eksfin) participation?

  Nscale press release — $790M Narvik financing, 11 May 2026

• Track Trump–Xi summit outcomes for AI chip and Hormuz read-throughs. The 14–15 May Beijing summit is the single most consequential near-term policy event for AI data center investors. Document what the summit produces (or fails to produce) on H200 export policy, AI safety dialogue, and any Iran/Hormuz framing. The range of outcomes is wide; the 30-60 day window from Issue 008 now has a named event attached to it.
• Commission transformer procurement audit across the US data center development portfolio. Presidential Determination 2026-10 means this is no longer a background operational item; it is a government-acknowledged critical path risk. Any development asset without a confirmed transformer purchase order and delivery date is carrying an unquantified construction schedule gap that may not be visible in current underwriting models.
• Validate demand signal independence for AI data center positions where Nvidia holds equity in the anchor customer. The circular investment concern does not require a thesis change, but it does require documentation: what proportion of contracted capacity is supported by commitments from entities in which Nvidia holds equity? The exit buyer will ask; the deal committee should have the answer before close.

• Develop Nordic and Baltic corridor data center exposure as a structural hedge. Nscale's $790M Narvik financing demonstrates that sovereign export credit participation, green power credentials, and geographic separation from Gulf and Hormuz risk zones are becoming investable at institutional scale. This is not a niche geography; it is an emerging institutional framework for non-Gulf European AI infrastructure.
• Map transformer and switchgear supply chain exposure to China across the full US portfolio and engage alternative supplier relationships now. The DPA invocation acknowledges a structural gap that cannot be resolved quickly; the $323M available DPA Fund is modest against $35B annual import dependency. Operators identifying allied-market alternatives (Korean, European, or emerging domestic suppliers) before the supply constraint tightens further will have a demonstrable advantage in capital raises and offtake negotiations.

• Build exit narratives that address the circular investment discount. As Nvidia's equity position in the AI infrastructure stack becomes better understood by the financial sponsor community, exit buyers will apply a discount to contracted capacity that cannot be validated as independent end-customer demand. Assets that can demonstrate organic, non-vendor-supported demand at scale will command a premium. Develop the evidence base before the exit process begins, not during it.