US Central Command said US forces struck five Iranian oil tankers after Tehran twice targeted an American warship over two days, hitting four Revolutionary Guards-linked vessels in the Gulf of Oman and a fifth near Kharg Island, Iran's principal crude export terminal. Iran responded with missiles fired at a US base in Jordan, 18 of 20 of which Jordan said it intercepted. The exchange is the sharpest escalation in a week, more than six months after the US and Israel first struck Iran on 28 February.
The forward signal sits in prices rather than the missiles. Brent crude rose 1.5% to $99.41 a barrel and US oil gained 1.6% to $94.55. Because Kharg Island moves roughly 90% of Iran's crude, the strikes raise a credible supply-disruption premium. A second front compounds it: Iran-backed Houthis attacked Saudi energy facilities on Tuesday, wounding 73 and temporarily halting operations including a 400,000-barrel-per-day refinery in Jazan, while Strait of Hormuz flows - previously about one-fifth of world oil - are nearly halted and the Houthis have declared a blockade of Bab al-Mandeb. Meanwhile safe havens sent a mixed message: gold December futures opened at $4,451.60 (down 0.6% on the day, but up 4.1% month-over-month), silver held above $66, and bitcoin fell 1.6% to open at $79,093.85 with ethereum down 1%.
Why this matters: The forward question is whether markets are pricing a geopolitical spike or a regime. Watch two things. First, the chokepoints: a 1.5% Brent move reflects a disruption premium, not a closure - if either Hormuz or Bab al-Mandeb is genuinely shut, the repricing would be an order of magnitude larger, and both are now contested at once. Second, the policy bind: a supply-side oil shock that lifts inflation while threatening growth is the textbook stagflationary trap, and a Fed leaning toward a hike into that shock is the scenario risk assets are least prepared for. The tell to monitor is correlation - gold and crypto falling together with equities would confirm the market is trading this as a rate shock, at which point diversification benefits thin out precisely when they are most needed. The prudent posture is to treat energy exposure as a hedge rather than a trade, and to watch this week's inflation print as the swing factor for the 15 September decision.
Cognition, maker of the coding assistant Devin, raised $2 billion at a $48 billion valuation in a round led by Andreessen Horowitz, Accel, Founders Fund, General Catalyst and Avenir. The valuation has nearly doubled from $26 billion just four months earlier, tracking a rise in annualized run-rate revenue from $492 million in May to $900 million now. TechCrunch read the round as a signal that investors see room for several major players in AI coding rather than a single winner.
The forward-looking comparison is with Cursor, which was in talks in April to raise at $50 billion on more than $2 billion of annualized revenue before selling to SpaceX for $60 billion - a deal driven largely because Cursor was severely compute-constrained. Cognition therefore now trades on a higher revenue multiple than Cursor did months ago. Management guides to $4 billion to $5 billion of annualized revenue by the end of 2026 and is training its own model on open-source foundations to cut its dependence on OpenAI and Anthropic. The offsetting pressure is cost: Cognition leases an Nvidia cluster costing hundreds of millions annually, potentially pushing cash burn toward $800 million this year.
Why this matters: For allocators tracking AI infrastructure, the signal to watch is where the value is migrating. The recurring lesson across both Cursor and Cognition is that the durable moat is compute capacity and the power and interconnect behind it, not the application interface. That argues for weighting exposure toward the infrastructure layer - GPU access, data-center capacity, energy - which captures value regardless of which coding assistant ultimately wins. On the venture side, a16z returning to lead a Cursor competitor after profiting from the Cursor exit suggests continued conviction in a multi-winner market, but the doubling of valuation in a single quarter on sub-$1 billion run-rate is the metric to stress-test: the forward risk is a valuation reset if revenue growth decelerates before the company reaches its own $4 billion to $5 billion target. The question to carry forward is whether Cognition can secure compute on better terms than Cursor could - because that, more than product, may decide its independence.
Sources: TechCrunch
OpenAI said an unreleased AI model, coordinating roughly 10,000 semi-autonomous agents, produced a solution to the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize problems each carrying a $1 million bounty from the Clay Mathematics Institute. The company said the run took about 88 hours and consumed 130 billion output tokens on that problem alone. Crucially, the result has not been independently verified or accepted by the Clay Mathematics Institute.
The claim is contested. NYU mathematician Tristan Buckmaster, working with Anthropic's Levent Alpoge using Codex and Claude models, announced his own set of proofs on Tuesday and alleged that OpenAI built on his group's unpublished progress, writing that information "had been passed to OpenAI" and that the company's answers on timing and human involvement "became more evasive." Across a week of work on several unsolved problems, TechCrunch estimates OpenAI consumed 300 billion output tokens, equivalent to roughly $22.5 million of compute at current rates.
Why this matters: The actionable read is on the cost curve of frontier AI. A single verified-or-not proof effort costing eight figures in inference is direct evidence that reasoning at scale is now gated by very large compute budgets - a data point that supports sustained demand for inference capacity and cuts against assumptions that per-token costs collapse toward zero in the near term. Watch two forward markers. First, verification: until the Clay Mathematics Institute or independent mathematicians confirm the proof, treat the capability claim as unproven and discount headlines accordingly. Second, governance: if frontier labs are perceived to be racing to publish on top of unpublished academic work, expect growing friction with the universities and grant bodies that supply talent and hard problems - a reputational and IP-provenance risk that is becoming a genuine diligence consideration for anyone investing in or partnering with the model layer.
Sources: TechCrunch, BBC
President Donald Trump issued executive orders on Tuesday banning a range of Canadian goods - alcoholic spirits, dairy and motorbikes among them - accusing Canada of discriminating against US businesses, with the ban set to take effect 29 September. It landed the same day Canada's dollar-for-dollar retaliatory tariffs came into force, with Prime Minister Mark Carney imposing duties of up to 50% on roughly $20 billion of US goods after talks collapsed in late August.
The trajectory points toward decoupling. Last month Washington placed 50% tariffs on about $20 billion of Canadian goods; Ottawa matched them on US steel, clothing and furniture. Now some products, including whey and non-alcoholic beer, are barred from US import entirely, while duties rise further on cheeses and motorboats. Carney told his country the pivot away from the US as its largest trading partner "will come at a cost," and no fresh negotiations are scheduled. The dependence is asymmetric: generally more than two-thirds of Canada's exports go to the US, while Canada is the second-largest US partner after Mexico but with a more diversified export base.
Why this matters: The forward implication for professional investors is twofold. First, inflation: tariffs and bans are a tax that flows into consumer prices, adding to the same price pressure that is shaping central-bank expectations - a factor to weigh alongside energy costs in the current macro read. Second, and more durable, is the supply-chain theme. When the world's most integrated bilateral trade relationship starts unwinding through bans rather than diplomacy, it accelerates friend-shoring and redundancy investment, supporting a structural rerating of domestic industrial capacity, logistics and near-shoring beneficiaries. The consensus view is that the economic pain forces an eventual deal; the risk worth hedging is that the dispute has acquired political momentum of its own, in which case what is being priced as temporary disruption is better understood as a lasting shift in North American trade architecture. Watch whether any talks are rescheduled before 29 September as the near-term signal.
NuScale Power jumped as enthusiasm around small modular reactors and AI-driven power demand continued to lift the nuclear cohort. The name has become a favored way to play the electricity needs of data centers. Momentum here reflects the growing link between energy and AI infrastructure themes.
Roivant Sciences surged, a large one-day move for the biopharma holding company. Sharp gains in the name typically follow clinical readouts, pipeline updates, or capital-allocation news. The move stands out within an otherwise mixed healthcare tape.
DigitalOcean rallied, drawing attention among smaller cloud and enterprise-software names. The cloud provider is increasingly viewed through the lens of its AI product push aimed at developers and small businesses. The move reflects renewed appetite for growth-oriented software.
Viavi Solutions climbed, a notable move for the network-test and optical-components supplier. Strength in the name often tracks optimism around network buildout and datacenter-interconnect demand. It offers a read on the optical and networking segment of the tech supply chain.
Novartis fell sharply, an outsized move for a large-cap European pharmaceutical name. Declines of this size typically stem from trial data, regulatory developments, or guidance concerns. The drop is worth noting as a signal for global pharma sentiment.
Howmet Aerospace declined, a notable pullback for a name that has been a strong performer within aerospace and defense supply. The engineered-components maker is tied to commercial aerospace build rates and defense demand. The move offers a read on profit-taking in the aerospace group.
Dyne Therapeutics, Inc. (DYN) traded at 20.31, -16.35% on the session.
Solaris Energy Infrastructure, (SEI) traded at 63.96, +16.29% on the session.
Geopolitics is doing the central banks' job for them this week. Fresh US-Iran military exchanges pushed Brent crude back toward $100 just as the Federal Reserve enters the run-up to its September 15 meeting, and with markets pricing roughly a 60% chance of a rate hike, the unusual prospect of tightening directly into an energy supply shock is now live. A widening US-Canada trade war adds a second inflationary impulse. Safe havens sent mixed signals: gold eased even as the metal held a month-over-month gain, silver stayed above $66, and Bitcoin slipped, a reminder that a higher-for-longer rate path is a headwind across both risk and non-yielding assets.
The frontier labs spent the week trying to convert raw compute into mathematical glory, and the messiness of that race was on full display. OpenAI claimed to crack a long-open aspect of a 90-year-old Millennium Prize problem in 88 hours by directing thousands of AI agents, only to face accusations from academic mathematicians that it built on their unpublished work. Beneath the headline breakthroughs, the more consequential story for enterprises is adoption pressure: firms are now ranking and even firing staff on AI usage, even as McKinsey finds 94% of companies have yet to see significant value. Meta, meanwhile, is betting the next phase is agents that act rather than chat.
Switzerland's early-stage engine stayed busy midweek, with the deeptech pipeline out of St. Gallen especially active. The University of St. Gallen's accelerator took in a fresh cohort, a St. Gallen AI startup pushed toward international expansion, and a Swiss drone maker showed the flip side of Swiss innovation by moving production abroad to scale.
Late-stage capital is concentrating in a handful of capital-intensive theses, with megarounds clustering in AI coding, nuclear energy and aerospace. The standout signal is Cognition's $48 billion valuation just four months after a $26 billion mark, evidence that investors still see room for multiple winners in AI coding rather than a single victor. Hard-tech is equally hot: reusable rockets, small modular reactors and pilotless cargo aircraft all pulled in nine-figure-and-up rounds, underscoring appetite for physical-infrastructure bets alongside software.