2026-08-20 Observatory briefing¶

60 days of a broken US-Iran MoU: the market stopped waiting for Hormuz¶

The Islamabad MoU’s 60-day window expired on 17 August with Tehran asserting a permit-and-toll regime over the Strait of Hormuz and Washington rejecting it, but the diplomatic failure was already priced into the physical market weeks earlier. The final week’s data shows 66% of Gulf of Oman net exports surfacing with no confirmed upstream, a reversal of the Week 3 peak when 95% of departing barrels could be traced to their terminal. The truce’s visibility dividend, the brief period when vessels loaded openly with AIS on, did not survive the truce.

The MoU solved one problem and left the other intact. Stranded floating storage fell from 61 mb at signing to 16 mb within three weeks, clearing the war’s backlog of laden tankers. But total crude on water inside the Gulf system closed the window at roughly 130 mb, above the war’s starting level of 96 mb. The ships got out; the system that stranded them did not change. Iran’s own loadings collapsed from 893 kbd in July to 156 kbd through 17 August, meaning Tehran ends the window exporting less than it did during active blockade.

The sequence of broken commitments is symmetrical: the US oil waiver survived 20 of 60 days, the blockade lift 27, and Iran’s mine-clearance obligation never began. Each side’s headline concession was withdrawn within a month, and each blamed the other. The Q4 bill is roughly 550 mb of crude shortfall against normal Hormuz flows, bridged so far by inventory draws that thin from September. Ballast entries into the Gulf have fallen to about two per day, so the market is not betting on a reopening. The strait is not closed by formal decree; it is closed by mines, war-risk insurance, and IRGC interdiction, none of which the MoU addressed.

A Fatal Strike and the First Seizure Since June as Iran’s Transit Rules Take Effect¶

Source: Hellenic Shipping News

The MINOAN DIGNITY’s chief engineer is dead, killed by a projectile that hit the engine room half a nautical mile off the Omani coast. The vessel had called at Bandar Imam Khomeini, loaded food cargo, and was exiting via the southern corridor. Its sister ship under the same Greek commercial management was struck on August 4 after the same port call. Iran’s targeting has been precise, aimed at Gulf energy companies and chartered tankers; a grain carrier is not that profile. The recurrence suggests a commercial dispute dressed as enforcement, though the evidence is circumstantial.

The August 17 seizure of the AMARA, a ballast tanker inbound to Jebel Ali, fits the stated terms of the expired 60-day US-Iran arrangement: transit authorisation and fee payment. Iran’s state media says the vessel violated maritime rules; AIS data shows a progressive slowdown, a loiter, a reversal. The absence of cargo weighs against a cargo-motivated interdiction. This is enforcement of a toll, not piracy.

The Chinese turnarounds are the sharper signal. Two state-linked VLCCs, one Hong Kong-flagged, aborted Hormuz transits within 24 hours, broadcasting "CHINESE CREW OWNER" in their AIS fields. COSCO and China Merchants, controlling over 100 VLCCs and roughly half of China’s Middle East crude imports pre-war, are keeping their fleets out of both Hormuz and Bab el-Mandeb on Beijing’s guidance. Iran’s assurances to friendly nations do not extend to their tankers’ safety. The market has stopped waiting for the MoU to hold; the 20.7 million barrels of laden crude waiting off Kharg Island and the reactivated Sulphur and LPG terminal after 23 vacant days are the physical record of that decision. The 118-kilometre oil spill from a dark VLCC in the waiting area is the environmental cost of a fleet that cannot move.

Trump vows ‘economic warfare’ on countries helping Iran¶

Source: Al Jazeera

Trump’s threat of “economic D-Day” against any country doing business with Iran is less a policy than a posture, and the vagueness is the point. By refusing to name targets or specify mechanisms, he converts the entire apparatus of US trade and finance into a standing threat against anyone who might test the sanctions regime. The message is aimed as much at European firms and Asian refiners as at Tehran itself: the extraterritorial reach of US law, from OFAC designations to dollar-clearing restrictions, already functions as a weapon against third parties. This merely makes the threat explicit.

The framing of “economic warfare” as an alternative to military action obscures how the two have become fused. Sanctions are not the opposite of war; they are war conducted through the banking system, with the same goal of strangling a state’s capacity to reproduce itself. Iran’s economy, already battered by decades of sanctions, faces the prospect of further isolation precisely because the US retains control over the global financial infrastructure. That control is the real content of the threat, and it is why the rhetoric matters despite its emptiness.

What remains unspoken is the contradiction in Washington’s position. The US demands that other states abandon trade with Iran while simultaneously insisting on the legitimacy of its own unilateral actions. For countries like China and Russia, which have deepened energy ties with Tehran, the threat is a reminder that the dollar system is a political instrument, not a neutral medium. Whether they can build alternatives to it, through bilateral currency swaps or barter arrangements, will determine whether Trump’s “economic D-Day” is a genuine turning point or just another escalation in a long-running siege.

Global Economic Convergence Is Stalling¶

Source: Project Syndicate

The IMF’s own data, marshalled by Keun Lee, shows the BRICS economies stopped closing the gap with the G7 around 2016. The arresting detail is that this is not a story of Western revival. The advanced economies, the US excepted, have been stagnating too. Convergence has stalled because the whole system has slowed, not because the hierarchy has reasserted itself.

Lee’s explanation centres on the composition of growth. The early catch-up phase, he argues, was driven by manufacturing and exports, the classic late-industrialisation path. That engine has sputtered. The frontier has shifted to intangible assets, digital platforms and finance, domains where the incumbent powers hold structural advantages in intellectual property law, standard-setting and capital markets. The emerging economies that thrived on absorbing existing technology now find the returns to that strategy diminishing, while the advanced economies find their own productivity gains increasingly captured by a narrow layer of rentier firms.

What is missing from Lee’s account is the other side of the ledger. The post-2008 period was defined by extraordinary monetary expansion. Cheap dollar credit flowed into emerging markets, inflating asset prices and corporate debt long before the growth slowdown became visible. The stall he dates to 2016 coincides with the first serious tightening of that credit cycle. The BRICS’ deceleration is not simply a technological plateau; it is the moment the fictitious capital that had papered over stagnant real accumulation began to be withdrawn. The G7’s relative stability, meanwhile, rests on the dollar’s exorbitant privilege, the ability to export inflation and import stability.

The rivalry between Washington and Beijing sharpens the picture. The US is not merely defending its position; it is actively weaponising the choke points of the global economy, from semiconductor fabrication to shipping lanes. This is inter-imperialist competition in its modern form, where the battle is less over territory than over the technological rents that determine the distribution of global surplus value. For the middle-tier economies, the space to manoeuvre between the two blocs is shrinking. The convergence that did occur was a product of a particular historical window, when US capital was willing to offshore production and China was willing to absorb the world’s excess savings. That window has closed, and Lee’s data is the tombstone.

Tariff pressures test trade alliances¶

Source: Hellenic Shipping News

The Peterson Institute's twin briefings arrive at a familiar juncture: the US executive branch, having exhausted its emergency powers and balance-of-payments justifications, now reaches for forced labour rhetoric to keep a blanket tariff regime alive. Alan Wolff's legal prognosis is blunt, and the coalition of 25 states plus affected businesses gives it weight. The tariffs' uniformity is their undoing in court and their tell in substance: sixty countries treated as one, regardless of actual labour records, because the policy's purpose was never remediation. It was revenue and leverage, dressed in humanitarian cloth.

Monica de Bolle's analysis of Mercosur shows the external pressure doing its work. Brazil and Argentina's bilateral deterioration predates Trump, but the tariff regime converts diplomatic friction into structural strain. A bloc built on collective negotiation finds its two anchors at odds precisely when Washington's protectionism and Beijing's competing overtures demand a unified front. The "potential implosion" she flags is not hyperbole; it is the logical endpoint of a strategy that treats allies as targets and then wonders why regional blocs fragment.

The two analyses share a premise worth stating plainly: the tariff weapon has outlived its usefulness as a tool of statecraft. It now generates legal defeats at home, diplomatic erosion abroad, and uncertainty for every manufacturer and carrier moving goods through affected lanes. For shipping, the practical consequence is a trading environment where route structures and charter decisions hinge on litigation calendars and bilateral spats rather than commercial fundamentals. The courts may strike the measures down, but the fragmentation they accelerated will not be reversed by a ruling.

A350F test aircraft emerges in distinctive ‘parcel’ paint scheme¶

Source: FlightGlobal

Airbus has rolled out the first A350F test airframe in a livery that mimics a courier parcel, complete with tape strips and address labels. The branding is a knowing nod to the freighter’s intended customer base, but the timing matters more than the paint. With certification flights due to begin shortly, the programme is moving into its most capital-intensive phase just as the air-cargo market shows signs of softening after a pandemic-era boom.

The freighter conversion race has always been about capturing the e-commerce surge, and Airbus’s decision to launch the A350F was a direct challenge to Boeing’s 777F dominance. The parcel livery is therefore not decoration; it is a statement of intent aimed at the integrators and express carriers who have driven demand for widebody freighters. Yet the underlying economics are shifting. Yields on major trade lanes have normalised, and the speculative ordering that characterised the 2021-2022 peak has given way to more cautious fleet planning.

What the paint scheme cannot conceal is the structural pressure on both manufacturers. Boeing’s 777-8F has faced repeated delays, handing Airbus a window, but the A350F’s success depends on converting launch orders into repeat business. The test aircraft’s emergence signals confidence, but the real test will come when airlines must choose between new-build freighters and converting ageing passenger aircraft, a cheaper option that grows more attractive as interest rates stay elevated. For now, the parcel livery is a promise. Whether the market delivers on it is another matter.

Avion Express provides A320s for FlyOne Armenia and Tarom¶

Source: FlightGlobal

Avion Express is stationing A320s in Yerevan and Bucharest for the summer season, adding two new customers to its wet-lease roster. The arrangement is a familiar one in European aviation: smaller flag carriers like FlyOne Armenia and Tarom lack the fleet depth to meet seasonal demand, so they rent aircraft with crews from an Irish-based operator that has built its business on precisely this kind of short-term capacity gap.

The timing matters. Summer 2026 is shaping up as another period of intense demand pressure across European leisure routes, and the carriers taking on these aircraft are doing so at the peak of the yield curve. Wet-leasing is the aviation equivalent of hiring temporary labour: it lets a carrier capture revenue it could not otherwise serve, while shifting the capital risk of aircraft ownership onto the lessor. For Avion Express, the model is attractive precisely because it avoids the long-term commitment of fleet planning. The company can chase the seasonal spike, then redeploy its assets elsewhere when demand softens.

What is worth noting here is the structural position of the two customers. Tarom, Romania's state-owned carrier, has been through repeated restructuring cycles and operates with a fleet that is small by regional standards. FlyOne Armenia, a relative newcomer, is competing in a market dominated by larger Russian and Gulf carriers. Both are using wet-leases to postpone the decision to invest in their own aircraft, a rational choice when interest rates remain elevated and the secondary market for narrowbodies is still tight. The arrangement lets them appear responsive to demand without committing the capital that would signal a genuine long-term strategy.

The broader pattern is one of consolidation by the back door. As smaller carriers increasingly rely on third-party capacity, the distinction between an airline and a sales operation blurs. The aircraft, the crews, the operational risk all sit with Avion Express; the flag carrier contributes little more than its brand and its route licences. That is a workable commercial model, but it hollows out the very thing state ownership is supposed to preserve.

Taiwan’s first F-16 Block 70 fighters seen departing Lockheed factory¶

Source: FlightGlobal

A planespotter in Fort Worth has photographed two F-16 Block 70s landing at Lockheed’s F-35 production facility, the first of 66 such jets ordered by Taiwan in 2019. The sighting confirms delivery is running ahead of a schedule that has already slipped once, with the aircraft due to arrive on the island by late 2026.

The Block 70 configuration is essentially a stopgap dressed as an upgrade: an airframe designed in the 1970s, fitted with an AESA radar and conformal fuel tanks to extend range. That Lockheed can still sell this as a front-line platform in 2026 speaks to the atrophy of the US fighter industrial base, where the F-35’s production problems have left a gap that only an ageing design can fill. Washington’s answer to Taiwan’s air defence needs is a refurbished Cold War workhorse, built in Texas and flown across the Pacific.

The delivery schedule matters less than the production line itself. Lockheed’s Fort Worth facility is the same site where F-35s are assembled, and the sighting of F-16s landing there is a reminder that the two programmes share infrastructure, tooling and skilled labour. Every Block 70 built for Taiwan is a small diversion of capacity from the F-35, which remains the Pentagon’s priority. The Taiwanese order is profitable, but it is also subordinate: a way of keeping the line warm and the workforce employed while the more lucrative programme sorts itself out.

For Taipei, the jets arrive at a moment when the balance of forces across the strait has shifted decisively. Sixty-six F-16s, however upgraded, do not alter that arithmetic. They are a political signal from Washington, priced in dollars and delivered on a schedule that suits Lockheed’s production planning more than Taiwan’s defence needs. The island gets a capable fighter, but the real transaction is the one between the US state and its defence contractors, with Taiwan’s security as the medium of exchange.

Stripe didn’t really buy OpenRouter because of the ‘singularity’¶

Source: TechCrunch

The Collison brothers' leaked letter reaches for the singularity as a joke, but the $7.5 billion price tag for OpenRouter is dead serious. Stripe paid nearly six times the startup's May valuation, and the founders alone bank $1.5 billion, more than the entire company was worth three months prior. That is not the arithmetic of visionary conviction; it is the logic of a toll booth operator buying the only bridge in town.

Stripe's core business collects a percentage of every transaction that flows through it. OpenRouter sits at the junction where developers decide which AI model answers their prompts, and crucially, where they pay for it. By owning the router, Stripe inserts itself into the capital flow before the money even reaches the frontier labs. The PitchBook analyst quoted in the piece has it right: this is about embedding into the middle of AI-era capital flows, not about human-machine fusion.

The competitive scramble matters here. Databricks wanted OpenRouter too, and Rippling and Ramp are building their own AI expense gateways. Every major platform is racing to become the metering point for AI consumption, because whoever tracks and bills the tokens controls the relationship with the developer. Stripe's acquisition gives it leverage over the hyperscalers and neoclouds that host the models, a position of intermediation that generates rent without producing anything itself.

The founders' talk of an economic uptick from AI is true in a narrow sense: more startups mean more Stripe fees. But the deeper logic is defensive. If AI agents begin transacting autonomously, the payment layer becomes the choke point. Stripe is not betting on the singularity; it is betting that whoever owns the payment rail owns the future economy, and it is willing to overpay to make sure that rail runs through its own ledger.

OpenAI seeks to one-up Anthropic with new customer privacy protections¶

Source: TechCrunch

The rivalry between OpenAI and Anthropic has moved from model benchmarks to the terms of data custody, and the shift is revealing. OpenAI's new Private Safety Processing, previewed to select customers, promises automated abuse monitoring across multiple conversations while retaining none of the customer's data. This is a direct counter to Anthropic's July policy of holding all sessions on its "covered models" for 30 days, a retention window that has spooked enterprises handling sensitive material.

Both companies are selling the same thing: surveillance without liability. Anthropic keeps the data but wraps human review in tamper-proof logs and "controlled access paths." OpenAI keeps nothing but reserves the right to receive a "narrowly defined signal" and then contact the customer for context. The customer ends up policing themselves either way, asked to volunteer the very data the company claims not to want.

The competitive timing is not incidental. Anthropic's annualized revenue run rate reportedly sits at $65 billion, and OpenAI's Q2 growth has slowed relative to its rival. Privacy policy has become a product differentiator precisely because the underlying product — enterprise AI access — has become commoditised. When the models are broadly comparable, the terms of tenancy become the battleground.

What neither company addresses is the structural position of the enterprise customer. The choice between 30-day retention and zero retention is a choice between two forms of dependency on a platform that defines what counts as abuse. The "narrowly defined signal" remains opaque, and the enforcement decision rests entirely with OpenAI. Enterprises are not being offered privacy; they are being offered a more palatable version of the same subordination, with the added burden of self-disclosure when flagged. The competition between the two labs may drive incremental concessions, but the underlying asymmetry between platform and tenant remains untouched.

How AI Could Hollow Out the U.S. Military¶

Source: Foreign Affairs

The Pentagon’s own doctrine assumes a stable human core: soldiers who judge, commanders who decide, and a bureaucracy that certifies both. Probasco’s argument is that AI dissolves that core from the inside, not by replacing people with machines but by degrading the cognitive capacities the entire edifice rests on. The evidence she marshals is genuinely unsettling. Computer scientists and oncologists given AI assistance performed worse at their jobs after it was withdrawn than before they ever used it. The skill was not augmented; it was atrophied. Automation bias, long a known hazard of navigation systems and cockpit instruments, becomes something more systemic when the tool is a generative model that produces confident, plausible output on demand.

The article’s sharpest observation is buried in its institutional detail. The Department of Defense’s autonomous weapons guidance was last updated in 2023, before Anthropic had released its first commercial model. Claude has since gone through twenty versions and developed superhuman hacking capabilities. The policy cycle operates on a timescale of years; the technology iterates on a timescale of weeks. This is not a failure of bureaucratic diligence but a structural mismatch between the military’s procurement and training apparatus and the commercial AI sector’s release cadence. Flying officers back for training on tools that change biweekly is, as Probasco notes, technically possible and practically absurd.

The deeper problem is that the military’s response to this mismatch will likely compound it. Staff reductions justified by imagined AI efficiencies mean fewer humans to exercise the judgment that AI purportedly supports. The exhausted sailor reviewing threat imagery is more likely to accept the computer’s verdict precisely because the human system has been hollowed out to accommodate the machine. The loop the Pentagon insists on keeping humans in is being widened by the very technology it is meant to contain.