The explosions started before dawn. By 9 a.m. Kyiv time, the city's military administration had a number: nine dead, thirty wounded, four of them children, across five districts. Seven of the dead were in Darnytskyi, on the left bank, where a residential block took a direct hit. In Solomianskyi, rescuers walked thirty-five people down from the upper floors of a five-storey building that was still standing but no longer safe.

This is the fourth summer of the war, and the arithmetic of these mornings has become brutally simple. Russia is firing ballistic missiles — Iskanders, Kinzhals — that travel at several times the speed of sound and give a defender roughly a minute of warning. The only thing in Ukraine's arsenal that can reliably kill a ballistic missile in flight is a Patriot interceptor, and Ukraine is running out of them. The United Nations counted June as the deadliest month for Ukrainian civilians since April 2022. July did not improve.

So when President Volodymyr Zelensky flew to Washington this week, he did not ask for tanks or planes. He asked for a license to build Patriot interceptors on Ukrainian soil. President Trump, by Zelensky's own account, said the two would have to be "very careful" about that. On Friday Zelensky reported a "good" follow-up call with Vice President JD Vance. On Saturday, nine people in Kyiv were dead. The gap between those two sentences is the entire policy.

What makes this morning different from the dozens before it is where the missiles are landing. Two days earlier, during a barrage that killed at least eight across western Ukraine and reached as far as Lviv, a Russian Kh-101 cruise missile crossed into Polish airspace and detonated in a field near Tarnawa-Kolonia, roughly 100 kilometers from the border. It left a wide crater and no casualties. It also left NATO scrambling F-16s, an airborne early-warning plane, and an aerial refueling aircraft over the territory of a member state.

Polish Prime Minister Donald Tusk rushed to the crater and chose his words with a diplomat's care: "There is no reason to think the target was Poland." His foreign minister noted that the Kh-101 is used exclusively by Russia's strategic aviation forces. NATO Secretary-General Mark Rutte called it "another reckless act by Russia and a dangerous consequence of its war against Ukraine." Nobody in Warsaw or Brussels used the word that hangs over the incident — Article 5 — because everyone understands that the moment you say it out loud, you are committing to what comes after.

This is the condition the war has entered: not escalation by decision, but escalation by leakage. A missile aimed at Lviv ends up in Poland because a Ukrainian pilot, out of interceptors, tried to shoot it down near the border and the geometry failed. The Baltics and Romania have logged the same intrusions for months. Each one is deniable, each one is small, and each one moves the frontier of the war a few kilometers west while everyone agrees it was probably an accident.

Inside Ukraine, the command structure is also shifting under pressure. Two weeks ago Zelensky fired Commander-in-Chief Oleksandr Syrskyi after six days of street protests — protests that began not over strategy but over the removal of Defense Minister Mykhailo Fedorov, the 35-year-old former tech entrepreneur seen as the engine of the army's drone-and-software modernization. Zelensky replaced Syrskyi with the younger Mykhailo Drapatyi and offered Fedorov a senior post. Fedorov, as of this week, has refused anything short of returning to Defense. A country fighting for its air is also fighting over who runs its war.

Meanwhile Ukraine is not only absorbing strikes. It has escalated its own campaign against Russian logistics, burning warehouses belonging to Wildberries, Russia's largest online retailer — the latest, in Volgograd, caught fire Friday. Kyiv says these are supply hubs for the Russian war economy. Moscow says its air defenses destroyed 274 Ukrainian drones overnight.

The shape of the next phase is already visible in these numbers. The war will be decided less by territory than by the sky over it — and by whether the interceptors arrive before the funerals do. As Foreign Minister Andrii Sybiha put it Saturday: the stronger the shield, the closer peace. Right now the shield is thin, and the missiles are landing in other people's fields.

For a few weeks in June it looked like the 2026 Iran war might actually stop. A memorandum of understanding, negotiated over agonizing weeks, put a floor under the fighting. Both sides now agree it is dead, and neither seems interested in mourning it.

The past week reads like a map of a war that has abandoned the idea of a battlefield. On Wednesday, US Central Command said it struck dozens of Islamic Revolutionary Guard Corps targets inside Iran. On Friday, Iran announced it had retaliated by targeting US military assets at Kuwait's Ahmad Al-Jaber airbase. In between, Saudi Arabia — for the first time since the war began on February 28 — joined the United States in striking Iran-backed militias in Iraq, killing at least twenty fighters of the Popular Mobilization Forces. Riyadh insists it acted in self-defense, and pointedly did not touch Iranian territory. The distinction matters to diplomats and to no one holding a rifle.

The geography keeps widening. Iran fired ballistic missiles at US installations in Jordan, which the US says it intercepted. A drone attack set two ships ablaze at Egypt's Damietta port — one of them a US-owned tanker — pulling Egypt into the war for the first time. Al Jazeera Arabic reported a drone strike on an Iranian Kurdish opposition camp east of Erbil. And in a genuinely strange convergence, a Ukrainian drone struck an Iranian commercial ship in the Caspian Sea, prompting Tehran to summon Ukraine's chargé d'affaires — the two largest wars on Earth briefly touching hulls in the same water.

Analysts keep naming the same tripwire. "The moment to watch," one told Al Jazeera, "is a direct attack that prompts a collective Gulf military response against Iran, as well as the threat that the Houthis escalate in the Bab al-Mandeb." The Houthis have already floated fees on commercial vessels transiting that strait, which is another way of saying they are pricing the world's shipping by the barrel.

What is notably absent is a theory of how this ends. A decisive US military victory over Iran, CNN's regional analysis noted, would likely require ground forces and a far larger campaign than anyone in Washington has appetite for. Mediation efforts led by Qatar and Pakistan exist mostly as press releases. The war has settled into a rhythm of strike and counter-strike across seven countries, each blow calibrated to hurt without quite forcing the other side's hand.

That calibration is the danger. A war fought entirely in the margins — a tanker here, an airbase there, a proxy militia somewhere else — has no natural stopping point, only an accumulation of incidents. Sooner or later one of them lands wrong. Ask Poland.

Start with the number, because the number is the story: $1.3 trillion in market value wiped from the world's twenty most valuable chip stocks in a single week, by CNBC's count. Nvidia alone lost $238 billion since Friday's close. SK Hynix shed $176 billion, Samsung $173 billion, TSMC $119 billion, Micron $113 billion, AMD $110 billion. Sandisk dropped 14 percent in a day.

Here is what makes this selloff different from an ordinary correction: it is not about demand. AI companies are still buying every chip that fabs can produce; memory supply is so tight that consumer RAM and SSD prices are rising and Apple has had to adjust. The market is not worried that the orders will dry up. It is worried about where the orders are coming from.

The word doing the work is *circular*. Nvidia is reportedly in talks to guarantee up to $250 billion of OpenAI's lease on a $500 billion, 10-gigawatt Ohio data center, plus a separate $350 billion to finance OpenAI's chip purchases. Stack that on prior commitments and Nvidia has something north of $750 billion of AI infrastructure in motion — much of it structured as Nvidia backstopping the customers who then buy Nvidia's chips. The concern writes itself: if the company selling the shovels is also lending its customers the money to buy them, how do you tell real demand from an accounting loop?

Jensen Huang calls the worry "ridiculous." The market's own instruments disagree with him. Nvidia's five-year credit default swaps jumped the most on record this week — CDS being precisely the instrument designed to price a $250 billion contingent guarantee. When the cost of insuring a company's debt spikes on the same news that its CEO dismisses, the tie goes to the swaps.

The bull case, laid out crisply by one analyst at Investing.com, is that this is nothing like the Lucent and Nortel vendor-financing blowups of the dot-com era. Lucent financed customers because it was pushing product against *weakening* demand. Nvidia is doing it against a shortage — it does not need the deal to move GPUs. On this reading, Nvidia backstopping OpenAI is a signal from the most informed party in the chain: the company with the best visibility into compute demand has concluded the math works.

Maybe. But the counter-fact is hard to wave away: OpenAI lost $38.5 billion last year, per FT and Ed Zitron's reporting, and has committed to $750 billion of infrastructure spend through 2030. Google just posted its first-ever negative-cash-flow quarter on AI capex. Morningstar's chief equity strategist called the selloff "loss of confidence" rather than deteriorating fundamentals — but confidence is the fundamental when your valuation is built on cash flows a decade out.

The honest read is that both things are true. Demand is real *and* the financing is circular enough that nobody can cleanly measure how real. Wall Street spent two years asking whether AI would reshape the economy. This week it started asking a harder question: how many years until the reshaping shows up as earnings — and who is holding the paper if the answer is 'more than the debt schedule allows.'

The Federal Reserve held its benchmark rate at 3.50–3.75 percent this week for the fifth consecutive time, but the vote — 9 to 3, with three officials pushing for a hike — was the most divided of 2026. That split is the whole macro story in one line: nobody agrees on which fire to fight.

On one side, the oil shock. Brent jumped more than 7 percent to around $88–90 a barrel as fighting resumed between the US and Iran, reviving fears about the Strait of Hormuz and now the Bab al-Mandeb, where the Houthis are weighing transit fees. Higher oil feeds directly into headline inflation and, worse, into inflation *expectations* — and the 10-year Treasury yield has climbed to 4.68 percent from 3.97 percent before the war reignited. That is the kind of move that lifts mortgage rates and squeezes equity valuations without the Fed touching a thing.

On the other side, a cooling labor market. A Reuters poll projects the July payrolls report, due August 7, will show a soft +91,000 jobs and unemployment ticking to 4.3 percent. In a normal cycle, that is a cut signal. In this one, LSEG data show markets pricing a 64 percent chance of a *hike* in September, because the oil-driven inflation risk overwhelms the labor softness.

"Today's numbers may have put the Fed between a rock and a hard place," Morgan Stanley's Ellen Zentner said of an earlier weak report — a description that has only gotten more apt. Cleveland Fed president Beth Hammack framed it as balancing a softening job market against "broad-based inflationary pressures" and argued policy can hold "for quite some time."

Complicating all of it is the new Fed itself. Chairman Kevin Warsh has scrapped the forward guidance that anchored market expectations for two decades. Plante Moran's Jim Baird warns that the loss of a clear communications framework will amplify volatility around every data release, because there is "a little less clarity around where we're heading." The next two dates — payrolls on August 7, CPI on August 12 — will now move markets more than the meetings themselves. The Fed bought itself time this week. Geopolitics will decide the price.

The numbers arrived before most people had seen the film. *Spider-Man: Brand New Day* generated more than $72 million in domestic previews — some counts put it at $74.5 million — dethroning *Avengers: Endgame*'s $60 million and setting an all-time record. Advance sales cleared $200 million, also a record. Projections have it landing somewhere between $270 million and $350 million for the opening weekend and north of $900 million worldwide, which would make it the second-biggest global opening ever.

And here is the genuinely surprising part: it appears to deserve some of this. The film opened to a 91 percent critics score and a 98 percent audience score on Rotten Tomatoes, an "A" CinemaScore, and reviews calling it the best Spider-Man since Sam Raimi's run. Tom Holland's fourth solo outing has "inspired ideas, striking kinetic filmmaking, and genuine emotional conviction," one critic wrote. Twenty-plus years after Spidey first swung across a screen, the character remains the most reliable emotional engine Hollywood owns.

The interesting tension is buried in the fan reaction. Even inside the euphoria, you can find the complaint that recurs under every tentpole now: "I'm tired of a movie's only goal being to move a franchise forward. This was not a stand-alone movie." The record-breaking machine and the exhaustion with the machine are the same audience, showing up on the same weekend, feeling both things at once.

There is also a quieter industrial story. *Brand New Day* is doing this *without* IMAX, which is still holding screens for Nolan's *The Odyssey* into September. It is doing it at the tail end of July, dragging what is historically Hollywood's softest month — August — into record territory and handing the theatrical business a lifeline into fall. In India it crossed ₹100 crore in two days, beating Nolan and Ranveer Singh's *Dhurandhar* combined.

So the machine works. That was never really in doubt. What *Brand New Day* proves is narrower and stranger: that a fourth solo film about a character we have watched for two decades can still out-sell every original idea in the building, and that audiences will pay record prices for the comfort of a story they already know by heart while complaining, in the same breath, that they wish it were something else.

Boy George was supposed to play King Herod at the London Palladium's *Jesus Christ Superstar* from August 3 to 15. On Thursday, his manager Paul Kemsley announced he would not be appearing — a "managerial call," made "with the utmost respect for the producers," to let "the production remain the focus." Richard Armitage will reportedly take the role. The word missing from Kemsley's statement was the reason.

The reason was a song. Days earlier, George posted "We Will Dance Again" (Hebrew title "עוד נרקוד") captioned "Shalom" — a reggae-inflected track supporting Israel and attacking pro-Palestinian musicians. Its lyrics: "You say genocide, I say war / When you're attacked, that's what the army's for." It accuses artists "holding flags, mouthing like sheep" of "selective memory." It borrows the name of the 2024 documentary about the Nova festival attack. Within days it was gone from streaming, George was off the show, and Artists for Palestine UK had called it evidence of "a profound lack of empathy."

Here is the part that keeps getting lost in the boycott arithmetic: George told his followers the song was made with AI. "I call it 'AI for good,'" he wrote. "Truthfully, I would struggle to write this with a human!" He had said earlier this year that AI had changed his whole approach to songwriting.

That sentence — *I would struggle to write this with a human* — is the actual story, and almost nobody is treating it as one. Strip away the politics for a moment, which you can do because George's position is his own and he has doubled down on it. What is novel is a working songwriter announcing that on the subject he cared most about this year, he reached for a machine because collaborators — humans, with their own consciences and their own limits — would have been friction.

That is what "AI for good" turns out to mean in practice: the removal of the last person who might have said *are you sure?* The technology did not invent George's opinion. It did something subtler and worse — it let him skip the one part of making art where another human being gets a vote. The controversy is being fought as a fight about Gaza, which it also is. But the thing that will outlast this news cycle is the quiet confirmation that the machine is now available precisely for the songs you couldn't get anyone to write with you.

There is a particular comfort in the phrase "interceptor shortage." It sounds like weather. Like something that simply *is*, a fact of physics and factory throughput, no more anyone's fault than a drought. Nine people died in Kyiv on Saturday, and the explanation offered — reasonably, accurately — is that Ukraine has run low on Patriot interceptors, the only weapon it has that can kill a ballistic missile in flight.

But a shortage is a ratio, and ratios have two sides. There is the supply of interceptors, which is finite and slow to grow. And there is the decision about where they go and who is allowed to make more. This week President Zelensky did not ask Washington for a handout. He asked for a *license* — permission for Ukraine to manufacture Patriot interceptors itself, on its own soil, with its own labor. The answer, per Zelensky, was that everyone would have to be "very careful."

Careful about what, exactly? The interceptor is a defensive weapon. It kills incoming missiles, not Russians. A license to build more of them cannot escalate a war; it can only make a city harder to kill in. The caution being exercised is not military. It is political, and it is about leverage — the quiet understanding that a Ukraine that can defend its own sky is a Ukraine that negotiates on its own terms, and that dependence is itself a bargaining chip that Washington is reluctant to spend down.

I am not arguing that this calculation is stupid. Great powers manage clients through supply, and always have. I am arguing that we should stop laundering the calculation through the language of logistics. "We are short on interceptors" is a true sentence that hides a chosen one: "We have decided that the marginal Ukrainian city is worth less than the leverage we retain by keeping Kyiv on a drip."

You can see the cost of that framing in Poland's crater. When a defender is out of interceptors, its pilots make desperate intercepts near borders, and the geometry sends the debris — or the missile — into a NATO field. The interceptor shortage did not stay inside Ukraine either. It rarely does. Scarcity, once you've engineered it, has a way of escaping the country you engineered it for.

So here is the modest demand. If the policy is to ration Ukraine's ability to defend itself in exchange for diplomatic leverage, let the people who hold that policy say so in those words, to the voters who will absorb the consequences, and to the families in Darnytskyi who buried the difference this weekend. Call it what it is. A shortage is an accident. This is a choice.

Boy George called the AI that co-wrote his war song "AI for good." He meant it, I think, sincerely. And that sincerity is exactly why the phrase deserves to be dragged into the light and examined, because it is going to be everywhere, and it is almost perfectly designed to mean nothing.

"AI for good" is a phrase with no object. Good *for whom*, decided *by whom*, measured *how*? George's song was, in his own view, unambiguously for good — a defense of friends he holds dear. To Artists for Palestine UK it was a denial of mass suffering. The technology adjudicated none of this. It could not. It took a prompt shaped by one man's conviction and returned that conviction in polished form, minus the friction of any collaborator who might have flinched. The "good" was smuggled in before the model ever ran. The machine just laundered it into a finished product and a plausible-deniability tagline.

This is the move to watch, and it is spreading far beyond one pop star. "AI for good" is what you say when you want the moral credit of intention without the accountability of authorship. When a human writes something incendiary, we know where to send the argument. When "AI helped," responsibility diffuses into the tooling. George half-performed this himself: *I would struggle to write this with a human.* The human collaborator is reframed not as a conscience but as an obstacle — and the machine's chief virtue becomes that it never says no.

Notice the structural resemblance to the other AI story in this issue. Nvidia backstops OpenAI's demand; the circularity makes it impossible to tell real conviction from manufactured. "AI for good" does the same thing to authorship that circular financing does to demand: it makes the origin unauditable. You cannot separate what the person genuinely believes from what the tool made frictionless to say.

I am not a machine-breaker. Used with a name attached and a hand on the wheel, these tools are extraordinary. The problem is the specific rhetorical package — "for good" — because it pre-empts the only question worth asking. The honest version of George's statement is not "I made AI for good." It is "I used AI to say something I couldn't get a person to help me say." That sentence you can argue with. "AI for good" you can only nod at, which is the entire point of it.

Demand the object. Every time someone tells you their AI is for good, ask them to finish the sentence. The ones doing genuine good will finish it happily. The ones using the phrase as a shield will change the subject — and that tell, at least, no machine can fake for them.

The most remarkable thing about the protests that swept Kyiv this month is that they happened at all. A country under nightly ballistic bombardment, four years into an existential war, put people in the streets — not against Russia, but against its own president's personnel decisions.

The spark was the removal, on July 15, of Defense Minister Mykhailo Fedorov. Fedorov, 35, a former tech entrepreneur, was widely credited with energizing Ukraine's defense ministry, driving its drone-and-software modernization, and pushing an anti-corruption agenda. Zelensky said he was forced to choose between Fedorov and Commander-in-Chief Oleksandr Syrskyi, 60, whom protesters and reformers viewed as an emblem of Soviet-school command culture — the two men, Zelensky said, had stopped speaking entirely.

Six days of demonstrations followed. A pro-government media outlet, United24, paused publication to join. A deputy Air Force commander resigned. Signs demanded Fedorov back and Syrskyi gone. On July 21, Zelensky bowed to half of it: he fired Syrskyi and named the younger, reform-minded Mykhailo Drapatyi — a battle-hardened general associated with progress in Kharkiv and reportedly close to Fedorov — as the new commander-in-chief.

But the crowds got only half a victory, and they know it. Zelensky offered Fedorov a senior role — reportedly deputy prime minister for military innovation. Fedorov refused, saying he would accept nothing short of returning to Defense. Protests resumed after a brief pause, with turnout lower but demonstrators vowing to continue "indefinitely."

It is a startling display of civic muscle in wartime, and a reminder of what Ukraine believes it is fighting for. But it is also a governance crisis at the worst possible moment. A military changing its top commander mid-bombardment, over a reform faction the president can neither fully embrace nor safely discard, is a military distracted from the one problem that matters this summer: the empty interceptor racks over Kyiv.