TheChek News Flow

The $80 billion annualized sales figure is remarkable, as few of the companies existed or generated revenue before 2023. Anthropic can claim about half that amount, as its revenue recently surpassed OpenAI’s on the strength of its AI for coding tasks and other white-collar work.

The revenue figures for the two AI leaders are somewhat inflated, however, given that each of them shares a material amount with business partners. Anthropic shares a portion of its revenue with cloud computing providers, namely Amazon and Google, that resell its models to their customers. And OpenAI must share 20% of its revenue with Microsoft, an early backer, through 2030. This year, that revenue share to Microsoft could be $6 billion.

The 32 AI model or application startups not named Anthropic or OpenAI may not be growing as fast as the leaders. But they aren’t far behind in that regard and continue to hit new revenue milestones, largely through sales of subscriptions and some usage-based fees.

Since December, three of those companies crossed $500 million in annualized sales, joining coding app Cursor in that club, according to our analysis. The three newbies are search firm Perplexity, voice AI provider ElevenLabs and coding app Cognition. (Our analysis doesn’t include infrastructure startups that rent out Nvidia AI servers.)

As with our past analyses of AI startup revenues, there’s some double counting going on. The four aforementioned startups that are generating more than $500 million in annualized sales, as well as the other companies on the list, likely collectively pay OpenAI and Anthropic billions of dollars a year for models to power their products.

Almost all of the companies’ products are powered by Nvidia servers, helping the chip designer generate the vast majority of profits so far in the AI boom. Nvidia has partly returned the favor, investing capital in at least 13 of the 34 companies on our list, including about $40 billion in OpenAI and Anthropic combined. The capital helps the companies keep renting Nvidia servers.

Our analysis doesn’t cover public companies that are also generating revenue from AI applications and selling models they developed themselves. Microsoft, for instance, has generated at least $6 billion in actual revenue from selling Copilot-branded apps, powered in part by OpenAI models that it uses for free.

Google sells a lot of Gemini AI models and Gemini-powered features in enterprise apps, and AI answers appear to be boosting its core search engine and by extension the search-advertising business.

A dozen major enterprise software firms such as Salesforce and ServiceNow are collectively on pace to generate billions of dollars a year selling AI-powered features and agents in their products. They’re also major customers of Anthropic and OpenAI, both for internal use and to power the products they sell. These AI buyers have been enduring recent price hikes from Anthropic.

Despite their fast-growing revenue, the AI startups are burning huge amounts of cash—likely more than $30 billion on an annual basis from Anthropic and OpenAI alone, in large part due to AI model-training costs. While some of the startups on our list have at times operated profitably, most are not.

For instance, Cursor’s gross margin was negative 23% as of the quarter ended in January, which was unusual for a startup generating such high revenue. The company’s gross margin has since turned positive, but the results show the challenge of relying on OpenAI and Anthropic technology, especially as Anthropic raises prices.

The Gulf state didn’t say who launched the attack, which highlighted the risks of the war with Iran

A drone strike set off a fire near the United Arab Emirates’ nuclear-power station, authorities in the Persian Gulf state said, a sign of the growing risks as the Iran war drags on.

The U.A.E. didn’t say who launched the drone but said the strike caused a fire at an electrical generator outside the inner perimeter of the Barakah Nuclear Power Plant about 175 miles west of the capital, Abu Dhabi.

Radiation levels were normal, and the nuclear station—where four reactors supply 25% of the Emirates’ power needs—continued to operate, the Abu Dhabi media office said. The United Nations’ nuclear watchdog, the International Atomic Energy Agency, said one of the reactors was on emergency diesel-generator power following the strike.

The agency’s director, Rafael Grossi, expressed grave concern about the incident and called for maximum military restraint around nuclear plants.

The attack comes amid fresh threats by President Trump to take military action if Iran doesn’t agree to a peace deal.

Diplomats continue to work on a deal, with Iranian media saying Tehran had recently received a new U.S. proposal for ending the conflict and had given its reaction to Pakistan, a key mediator in the conflict. There was no immediate response from the White House on the Iranian media reports.

The two sides are far apart on issues like Iran’s nuclear program and the extent to which it should have any control over the Strait of Hormuz. Trump said a week ago that Iran’s previous response had been “totally unacceptable.”

Iranian state media listed the nuclear plant among Iran’s targets for retaliation after Trump threatened in late March to obliterate the country’s power plants starting with the biggest one if Tehran didn’t lift its chokehold on the strait.

Trump hasn’t carried out the threat, and Iran continues to bottle up the strategic waterway. Iran, however, has continued to target the U.A.E. and has sharpened its threats after the Gulf state said it would strengthen its security ties with the U.S. and Israel to counter the threat from Iran.

The U.S. and Israel carried out widespread strikes on Iranian nuclear sites since last June, including in this year’s fighting, although none of them caused any kind of radioactive fallout. The strikes included attacks very close to Iran’s Bushehr nuclear reactor.

Iran, backed by Russia, has repeatedly called for international condemnation of strikes on nuclear facilities.

The U.A.E. has carried out military strikes on Iran since early in the war, hitting targets including a crude-oil refinery on Lavan Island in early April, in coordination with the U.S. and Israel, The Wall Street Journal reported earlier.

The Barakah plant was built under a 2009 agreement with the U.S. that allowed the transfer of peaceful nuclear technology in exchange for U.A.E. agreements to allow inspections and import its fuel rather than enrich or reprocess uranium domestically.

Plus, what to look for in a used EV, Anthropic’s AI lead, the Cerebras IPO, the OpenAI lottery tickets and the most AI-proof jobs in tech

Both sides gave it their best shot. Now it’s up to a jury—and U.S. District Judge Yvonne Gonzalez Rogers—to determine a winner in the titanic fight between AI kingpins Elon Musk and Sam Altman.

The world’s richest man jetted off to China as President Trump’s restored “First Buddy” while the third and final week of his trial against OpenAI and cofounder Altman played out in an intense fourth-floor courtroom in Oakland, Calif. There, the diminutive Altman testified, against one Musk lawyer’s portrayal, that he was not in fact a liar.

The billionaire also defended converting OpenAI from a nonprofit to a commercial entity—a move that sparked the lawsuit by Musk, a onetime major donor, alleging breach of charitable trust and unjust enrichment—as essential for continued growth.

Altman skipped the final day of testimony Wednesday, missing a welcome comedic break when the judge kicked things off with a lawyerly debate on whether to admit one of the most talked-about items in Silicon Valley: a trophy of a golden donkey’s hind quarters, awarded to an OpenAI intern after Musk called him a “jackass” during an argument over safety. (The intern is now OpenAI’s chief futurist.) Gonzalez Rogers held the statue with mock, or perhaps real, disgust.

After closing arguments Thursday, the next move is jury deliberations Monday.

In 1950, Robert Doisneau photographed a kiss you’ve surely seen. The man and woman seem to have been stopped by ardor amid the midday rush, in front of the Hôtel de Ville, Paris. Though staged, it became, nearly immediately, one of the most iconic images of the 20th century.

Surely unknown to Mr. Doisneau, nine years earlier, there was another kiss captured on film in Paris that was much more spontaneous, just as passionate — far more desperate — between two Jews about to be separated by Vichy police. This kiss, found on a contact sheet of Nazi propaganda images in a Reims flea market six years ago, is now at the heart of a new exhibition of 98 Nazi propaganda photos at the Shoah Memorial in Paris curated by Lior Lalieu and me. This kiss, perhaps destined to become just as iconic, reveals a very different midcentury Parisian moment.

The photos provide a detailed visual account — almost minute by minute — of the very first, and little known, roundup of Jews in France on May 14, 1941. That day, some 3,700 foreign-born Jews obeyed a summons by Paris police with a notice, printed on light green paper (it became known as the “green ticket roundup”), for what they believed would merely be a check of their immigration and identity papers. The operation was organized by a man named Theodor Dannecker, the envoy of Adolf Eichmann in Paris. A photographer with the Nazi propaganda unit in the city was on hand to observe.

What gives these newly discovered photographs their singular power is not only what they show but the fact that they survived at all. They remind us that the past is never entirely buried, and that images can unexpectedly return to challenge the void of memory and representation. They function today not as propaganda, the purpose for which they were originally produced, but as fragments of truth — painful, incomplete and indispensable — that allow us to better understand the way the roundup was organized and conducted and also to get a glimpse of the victims’ shock, fear and pain.

There are only a few hundred photographs of roundups or murders of Jews from the 1930s and 1940s, a disparity of mass proportion considering the extent of the genocide. Some were taken by victims as acts of resistance, some by bystanders, and others, like this collection, were by an authorized photographer for the Nazi propaganda machine. From time to time, grandchildren of Nazi perpetrators find these images in attics and boxes when the older generation passes away.

This particular group of photographs was meant to document a Nazi success story. They begin with the trap: Jewish men and their spouses were summoned to over 60 locations in Paris: police offices, various administrations and a sports facility in the 11th Arrondissement. Women, we know from eyewitness testimony, were asked to return home to gather items; a list was provided. When they returned, as the images we now have on hand show, they were barred from reuniting with their male relatives. The doors were closed and guarded by French policemen. We can see the women’s pain, their bewilderment, many with bundles in their arms. We see couples as they part from each other.

Other photos document the departure of guarded buses, commandeered from the Paris bus company, filled with the captured men. We see the arrival of the internees at Paris’s Austerlitz train station. The same photographer captures the imprisonment of these Jews a few days later at the French internment camps Pithiviers and Beaune-la-Rolande. Some 700 of those Jewish men were later liberated, or escaped, after the green ticket roundup. About 3,000 of those taken that day were later deported to Auschwitz-Birkenau; of that group, just a scant few returned.

Taken months before the decision to annihilate the Jews of Europe was made, these 98 photos do not show extermination camps, gas chambers, shootings or even starvation. What they do show is the careful, methodical beginnings of racially motivated separation that later enabled the mass murder. The noted historian Raul Hilberg called this phase of the Nazi genocide “concentration.” There is no sign of outright violence; indeed, the despair of the ensnared Jews and their bewildered spouses is shown with a strange sensitivity by the German photographer.

At first the photos were kept on file by the German Propaganda Unit in Paris. After the war, six of the 98 photos were found in the archives of the N.I.O.D. Institute for War, Holocaust and Genocide Studies in Amsterdam, an indication that they were shared among the various propaganda units across Western Europe. A few others circulated among archives. But the vast majority were languishing, unseen, on contact sheets, until 2020, when two amateur collectors came upon them at a flea market. They brought the sheets to Ms. Lalieu, the director of photo collections at the Shoah Memorial in Paris, who analyzed the images in an effort to identify as many people as possible. Ms. Lalieu also identified the photographer as Harry Croner, a man from Berlin, who had gone on to have a stellar career in postwar West Berlin, as a famous cinema and opera photographer. (Half Jewish himself, he spent the end of the war in a labor camp.)

After the photos were found, Ms. Lalieu invited first and second-generation survivors to the memorial in an attempt to identify their parents or grandparents in the pictures. There were some extraordinarily moving moments. Though only five attempts at identification were successful so far, each marks a small victory against the backdrop of Nazi cover-up and of the looming loss of memory. A few of the photographs were shared publicly for the first time in 2021.

One year after these photos were taken, in mid-July 1942, about 13,000 Jews — mostly women and children — were rounded up, pulled out of Paris apartments and taken to an indoor sporting arena, the Vélodrome d’Hiver, in southwestern Paris. They received little food or water and were subject to abject conditions. Depleted and distraught, they were sent from there to the internment camps of Drancy, Pithiviers and Beaune-la-Rolande. Later, a majority were sent on to Auschwitz.

It was an arrest of mass proportions, a stain on French history. We know this history from eyewitness testimonies, from memoirs of the few survivors and from the mass of police documents. Only a single photograph of the roundup is known. In it five buses are parked alongside the Vélodrome d’Hiver. The image was most likely taken clandestinely from an overlooking window. With the discovery and exhibition of this new group of photographs, the picture and understanding of the Holocaust in France have deepened.

The value of an image is entirely dependent on context. Taken to prove racial superiority, these 98 photos on display through December now show depravity. They also jar awake memory, shore it up against time. In an ephemeral era of mass documentation — of our own lives, of our public and private existence — the re-emergence of these photographs is a tangible reminder that some images refuse to be erased from our collective past.

The stock-market surge has propelled the use of a new kind of tax-loss harvesting. We break it down.

With the stock market near record highs, it isn’t enough to be winning anymore. Wealthy investors are now obsessed with losing, too.

The hottest investment on Wall Street promises a magical-sounding mix of both: Index-beating performance that also comes with losses to offset capital-gains taxes.

Millions of investors are sitting on big gains in concentrated stockholdings, be they shares of Magnificent Seven tech giants, plain-vanilla index funds or employee shares of hot tech companies.

The traditional move would be to turn to direct indexing, a form of tax-loss harvesting that has been an investing mainstay since the early 2000s. There is now $1.1 trillion invested in that strategy, according to research firm Cerulli Associates.

But for many investors who have ridden this ebullient stock market, direct indexing is no longer generating the losses it once did.

So financial advisers are trying to turn up more losses for their clients by employing strategies that use leverage. Over $150 billion has flowed into this newer breed of tax-loss harvesting, known as long-short tax-aware, or tax-aware alpha, according to Brent Sullivan, who writes about the industry on his Substack, Tax Alpha Insider.

The strategy comes with some risks for investors.

Let’s start by looking at how direct indexing works.

If the S&P 500 is up 15% for the year, the goal is for the client’s portfolio to rise that much, while also distributing losses. These losses could be used to offset gains in highly appreciated securities elsewhere in the portfolio.

But direct indexing gradually loses steam as a method of generating losses as the market continues to rise. Around half of such portfolios industrywide are now “ossified,” meaning they are no longer generating losses, estimates Jon Diorio, head of U.S. wealth product at BlackRock.

A new strategy took shape. Hedge-fund firm AQR Capital Management developed the tax-aware long-short strategy for clients such as family offices and ultrawealthy individuals. Around five years ago, the firm and its rival, Quantinno Capital Management, began allowing individual investors and their advisers to tailor the strategy inside their own separately managed accounts, or SMAs.

Custodians, most notably Fidelity Investments, made the strategy accessible by lowering the investment minimums. It was a hit. AQR now manages nearly $70 billion in tax-aware long-short strategies, roughly 80% of which is in SMAs. Quantinno manages more than $48 billion. They have spawned numerous imitators.

Long-short managers generally aim for “alpha,” meaning they want to outperform the benchmark in addition to harvesting losses. The strategy must balance these aims.

The strategy has become so popular that Fidelity and Charles Schwab have recently instituted limits on new accounts.

Both strategies come with risks. Direct-indexing managers might not succeed in tracking the index if they can’t find appropriate substitutes for the stocks they sell. The Internal Revenue Service’s “wash sale” rule limits the ability to recognize a loss on the sale of stock if the same stock is purchased within 30 days before or after the sale.

Neither direct indexing nor long-short SMAs are likely to eliminate all capital gains in a bull market like this one. The losses they generate are often used to offset gains in other areas of the portfolio. For example, someone selling a stake in a business at the beginning of the year could invest the money into a long-short SMA to generate losses to offset some of their capital gains by the end of the year.

But holding on to the winners means new gains can build up over time within the strategy. Investors must eventually pay taxes on these if they cash out.

Michael Paulus, founder of wealth manager PCM Encore, recommends direct indexing for nearly all his clients, particularly those close to retirement who will need to sell highly appreciated stocks to fund it. Direct-indexing managers charge annual fees as low as 0.05%.

But he is more cautious about using long-short SMAs. He has used them for clients expecting a major one-time gain such as from the sale of a business or for clients who work in tech and get a large portion of their pay in stock. He otherwise avoids them.

Tax-aware long-short strategies can cost as much as 1.5% to 3%, including investment management, financing and borrowing fees.

“All else being equal, I’d probably rather cut the government a check than a hedge fund in Connecticut,” Paulus said.

AI Chip Mania Sows Seeds of Its Own Destruction

Investors already factor in cyclicality in the chip industry. The bad news is that they’ve frequently been wrong before.

Investing in artificial intelligence involves a strong belief that it’s different this time, and memory-chip makers are a particularly extreme example.

Micron Technology MU -6.62%decrease; red down pointing triangle recorded its biggest-ever loss just three years ago, and is now forecast to become the sixth-most profitable U.S. stock. It will make just under $100 billion over the next 12 months, more than Meta or Berkshire Hathaway. Micron and its rivals are big winners from runaway AI demand and soaring prices for the high-bandwidth memory Micron makes.

Like Samsung Electronics and SK Hynix, it’s in the sweet spot of the chip cycle, boosting prices, profits and their stock. It goes outside chips, too: Micron made a notable contribution to Wall Street’s upgrades of the S&P 500 earnings outlook, while the two Korean stocks have made the country’s market by far the world’s best-performing this year.

The question is how long booming demand for memory chips can last.

Memory chips are a perfect example of a highly cyclical industry. Heavy investment is required to build a fabrication plant, or fab. When demand rises, it takes several years for supply to catch up, during which prices and profits jump. Those high profits encourage CEOs to expand supply. And the high fixed costs encourage producers to run fabs at full capacity—even when supply overshoots demand. The cycle turns when excess supply pushes down prices and profits plunge, as they did in 2022-23.

Already the high profitability has encouraged heavy capital spending. Micron is spending $150 billion to build or expand fabs in New York, Idaho and Virginia, and new Korean fabs are opening.

The good news is that investors already factor in cyclicality. The bad news is that they’ve frequently assessed the cycle wrongly at key moments in the past.

The risk of a downturn is embedded in Micron’s valuation. Two weeks ago it was the S&P’s third-cheapest stock measured by price to forward earnings, and it’s still at under 10 times, tame for a highflying stock. That doesn’t make it cheap, though. It just means investors recognize that the boom times in memory chips never last.

History shows how this works. In the last cycle Micron stock peaked at the start of 2022, with the forward P/E at just nine times, ahead of a halving in the shares that year. The stock bottomed out and subsequently doubled after the loss was baked into predictions.

Something similar happened in the mid-1980s and 1990s cycles. When the stock peaked in 1984—at a level it took another nine years to surpass—it traded at only 15 times forward earnings. In the 2018 cycle the stock peaked at just 5.5 times. Losses for investors who were fooled into thinking they were buying a bargain were vast.

For now, any danger comes from demand, not supply. New capacity due this year and next isn’t enough to crush profits, so long as demand holds up. So what could go wrong?

The biggest risk is impossible to quantify: AI technology could become far more efficient in its use of memory, meaning data centers need less of it. Memory stocks took a hit in March when Alphabet researchers published a paper showing dramatic improvements in memory efficiency, but have recovered. Large language models are an immature technology, and engineering improvements for specialized data centers should be expected—but how big they are and when they come is unknowable in advance.

Other risks apply to the whole AI supply chain: Data-center plans may be scaled back, AI uptake prove slower than hoped, or a political backlash may hinder expansion. All are plausible; none are considered that serious by the AI bulls driving stock prices.

A final risk is that supercharged profits attract new rivals to enter the market. For now, that seems unlikely in the superfast memory Micron makes, but it’s already happening with other highly profitable chips used in AI.

The economics of chip makers outside memory chips are similar, but as their products are more differentiated—think Nvidia versus AMD—they are much less cyclical.

Fat margins on Nvidia’s chips have persuaded Alphabet to develop what it calls tensor processing units, or TPUs, dedicated to training AI, one of the big uses of Nvidia’s expensive graphics processing units (GPUs). Amazon’s Graviton chips provide the central processing unit (CPU) for the “inference” involved in using AI models, demand for which has boosted Intel.

Recent entrant Cerebras, which launched the first of its giant chips for both training and inference only in 2019, raised $5.55 billion in its IPO on Thursday, and its shares more than doubled immediately.

While AI demand is soaring, all this extra supply can be absorbed without much effect on profit margins. But the longer it goes on, the more competitors enter, and the more capacity is built.

As with all commodities, success sows the seeds of its own destruction—even if AI hopes are fulfilled.

How Britain can tackle its mountain of unpaid energy bills

A long-term plan such as a social tariff could stop vulnerable households from running up debt

There are some things UK households pay for out of their own pockets — like food, petrol and holidays. Then there are things the government is expected to fund, such as schools and healthcare. Gas and electricity fall somewhere in between. And because of this ambiguity, the country is racking up unpaid bills with alarming speed.

The UK’s energy retailers — companies including British Gas and Octopus — are collectively owed £4.6bn in bills, according to energy regulator Ofgem, up from about £1bn at the beginning of 2018. The Iran war will raise household gas and electricity prices come July, adding to the strain.

This needs fixing, not least because the cost of non-payment is, in effect, socialised. Debt-related costs are factored into the price cap — the maximum price that the regulator allows retailers to charge their customers — weighing on the nation’s already expensive bills. But what the fix should be depends on the cause of the build-up, and there is more than one factor at play here.

Financial distress is clearly part of the story: one out of every four households currently in energy debt is at or below subsistence living levels, consultancy Baringa thinks, while a further two are considered “vulnerable”. Ofgem has proposed writing off about £500mn owed by those who really cannot pay, and adding the cost to everyone else’s bills.

It would be better to stop those who struggle from running up debts in the first place. Crucially, there should be a long-term plan, such as a social tariff, to make sure that the most vulnerable households can access heat and light. And this should be run by the government, rather than energy companies and their regulator.

The catch is that those who want to game the system have plenty of opportunity. There is some evidence that households are “prioritising other payments” as industry body Energy UK puts it. Giving suppliers some bite would help. Pre-payment meters are a useful tool, but having seen one of their own rightly walloped for inappropriately forcing them on vulnerable customers, they currently do not install many.

On top of that, a huge one-third of total energy debt has been built up by those moving house. Since disconnection is vanishingly rare, new residents can access the property’s gas and electricity without setting up an account. Sorting this out should be within the limits of human ingenuity: Ofgem is consulting on the idea of giving unnamed occupiers energy through pre-payment meters, switching to direct debit once a new account is set up.

Suppliers can do their bit to help, by billing more frequently and more accurately, for example. But as long as the UK faces a broader affordability crisis, some people will simply be unable to pay for essential services. That is a problem for government, rather than the energy sector, to solve.