Crisis at OpenAI — now they’re changing strategy

April 2, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on April 2nd, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

With ChatGPT, OpenAI launched the biggest tech trend in a decade. It looked unbeatable. But now CEO Sam Altman’s operation is shaking. What is happening inside the world’s leading AI company?

It is an unusual gathering of powerful figures standing in a row in the Roosevelt Room of the White House.

It is January 2025 and Donald Trump takes to the podium. Beside him stand two of the world’s most influential tech billionaires — SoftBank’s Masayoshi Son and Oracle’s Larry Ellison.

Furthest out stands the youngest and least wealthy of them all: Sam Altman, CEO of OpenAI. He looks a little uncomfortable in a dark suit and brown shoes.

But Altman has travelled from Silicon Valley and dressed up for a reason.

Together with the other men, he is about to present Stargate — a gigantic AI investment in the US, costing up to 500 billion dollars.

The scale was compared to the Manhattan Project, through which the first atomic bomb was developed in the 1940s.

A year later, it is clear that nothing of that calibre will materialise.

The men in suits could not cooperate and the Stargate project has stagnated. Instead of a joint investment in data centres, new partners have had to step in and build. OpenAI — which needs enormous amounts of computing power for its AI services — will end up renting it from others rather than building its own.

The reason for this is simple. OpenAI cannot afford it. And according to The Information, they cannot borrow the money either.

This, along with several recently abandoned projects, points to a new strategy for the AI industry’s market leader.

A radical shift appears to be under way.

Stargate is not the only project to have been forced into major changes quickly. In December last year, Disney invested one billion dollars in OpenAI and announced a collaboration involving the animation giant’s many characters. Through OpenAI’s video app Sora, users could make their own Disney videos — legally and with copyright secured.

Just 103 days later, the fun was over. On 24 March, OpenAI announced that it was shutting down Sora and several other video creation services.

Disney said it “respected” the decision to focus elsewhere, but it clearly did not look good. The agreement between them could have set the standard for AI and the film industry in Hollywood. Instead it became an embarrassing belly-flop.

There are two reasons for this rapid reversal — and both relate to money in different ways.

The first is about focus. According to the Wall Street Journal, OpenAI’s COO Fidji Simo told her staff they did not have time to be distracted by “side projects.” Admittedly, no one called Sora a side project when it was launched. But in contrast to what Simo wants the company to focus on, one can understand the categorisation.

OpenAI is now to concentrate on “productivity” — and enterprise productivity in particular. Companies that can pay their way.

The second reason behind Sora’s closure is pure cost-cutting. Creating video with AI is extraordinarily expensive and requires very large amounts of computing capacity. OpenAI can put that to better use elsewhere. Relative to how little money video as a category brings in, it is clear that Sora cost substantially more than it was worth.

Sora is not alone in being shut down. That list is long.

Together with e-commerce company Shopify, they were going to launch the ability to buy products directly within ChatGPT. That project has also been scrapped.

And the “erotic chatbot” that was promised has been on ice for a couple of days.

Many are also waiting for the hardware product that OpenAI is to release with Apple’s former chief designer, Jony Ive. According to information from American courts, it has now been pushed back to 2027.

Given the above, one might ask a simple but in this context unusual question. Is OpenAI short of money?

What makes it unusual is that OpenAI has raised around 168 billion dollars in various forms of financing. One might think that should be enough for most things.

As a comparison, that is more than the entire Swedish government budget for a year.

Despite this, there is much to suggest that the money is beginning to run out — or at least that the costs visible on the horizon will exceed revenues for a long time to come. The company’s own forecast points to losses of around 14 billion dollars — just this year. In total, OpenAI estimates it will take around 111 billion dollars — approximately 1,000 billion kronor — before it turns profitable in 2030.

Anyone who has ever seen, or made, such a forecast knows that at best it is an educated guess. Nobody knows what the real number might be.

But the strangest thing of all is how the company found itself in this pressured situation, and what it now plans to do differently.

To understand that, we need to go back a couple of years.

The icon showed a white “N” against a teal gradient. Double-click it and the entire world’s internet opened up for home users. The “N” stood for “Netscape Navigator” — the popular web browser of the day that let users surf the web. Something that was new at the time, in the mid-1990s.

In this world, Netscape quickly became dominant. The browser was released in 1996 and just two years later it was reported to have 38 million users. That made it the world’s most popular PC software ever, at that point in time.

With a market share of around 89 percent, Netscape looked unbeatable. Like OpenAI, they were overwhelmingly the largest — and were introducing a new paradigm of technology to the public.

The success would prove short-lived.

Despite the initial triumph, it took just two years before Microsoft’s browser, Internet Explorer, overtook it. That same year, 1998, Netscape had 452 million dollars in revenues but made a loss of 159 million dollars.

Compared with today’s AI figures, it almost sounds quaint. Just 159 million dollars in losses in a year?

But at the time, the loss resembled a black hole. The following year Netscape was acquired by AOL, and the decline just continued. By 2002 the market share was just above three percent.

The market’s overwhelming leader — those who introduced the concept of the “web browser” to the public — had fallen.

Netscape was consigned to the history books.

Even if OpenAI is not exactly like Netscape, there is a lesson to be learned.

Being first to market is a good start — but not necessarily more than just that. What is usually called “first mover advantage” is often fairly short-lived.

Even when it comes to entirely new categories of software, market shares can shift quickly. Those who were around early may also remember AltaVista — the undisputed biggest and best search engine of its day. Then Google came along. Everyone knows how that ended.

It seems as though OpenAI is beginning to grasp this. It is therefore more important than ever that the many billions they have raised in investment last longer.

They need to hold out against competitors and a changing market.

Standing in the way of OpenAI’s world domination are two siblings who were previously employed by Sam Altman: Dario and Daniela Amodei, founders of the AI company Anthropic and the chatbot Claude.

With its beige app and somewhat more talkative style, Claude has quickly become many people’s favourite AI tool. They have been around since 2021, when the Amodei siblings left OpenAI along with several colleagues.

Anthropic is now growing substantially at OpenAI’s expense. But this is due more to OpenAI making poor decisions than to Anthropic suddenly becoming so much better.

Take the recent Pentagon incident. When Anthropic refused to let the US Department of Defense use Claude for all conceivable purposes, it became a major conflict. The military threatened to classify the company as a “supply chain risk,” which would have pressured many others to stop working with them. Anthropic in turn sued the Trump administration.

Into the vacuum that emerged stepped Sam Altman and OpenAI. From the outside it appeared as though they agreed to everything Anthropic had opposed, though the exact details of what transpired are not known. Nonetheless, the signal value was enormous — had Altman’s sense of responsibility for AI safety disappeared? Many users accustomed to AI products have been able to switch easily to Claude.

Another reason for Anthropic’s sudden success relates to the focus that OpenAI COO Fidji Simo called for above. While OpenAI invested in AI-generated video and a long series of different projects, Anthropic focused on a specific segment of the market: enterprises — and those working with programming in particular.

The products Claude Code and Cowork have quickly become extraordinarily popular. Programmers have gone from writing code themselves to instructing and verifying the code that AI systems build.

Anthropic’s strategy is smart. In this customer segment one finds those who are most likely the fastest — and most eager — to use AI services.

The success was immediate. OpenAI has a similar product, Codex, with which one can programme. But it has been somewhat lost among all the other initiatives under way.

When COO Simo talks about focus, this is what she means.

OpenAI needs to become more like Anthropic.

It is not always easy being the market leader. The defectors at Anthropic are showing the way to enterprises. Google, after a weak start, has caught up considerably and has well-stocked coffers. The queue of companies wanting access to the attractive AI market is long.

Can OpenAI adapt in time? An IPO looms — and with it a new funding lifeline.

But Netscape was listed on the stock exchange too, back in the 1990s.

And you know what happened to them.

The game is rigged — and that’s the point

March 31, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 31st, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

Is it geopolitical analysis or gambling? Prediction markets have suddenly taken on a prominent role in geopolitics. But now questions are being raised about corruption — and military risks.

“Will American troops land in Iran before… 31 March?”

The question is not merely geopolitical — it is also the basis for bets worth around 200 million kronor. On the American prediction market Polymarket, bettors placed the probability of a “yes” at 29 percent about a week ago. By Monday it was down to 6 percent.

By making assessments about world affairs, war has been turned into a game. On prediction markets, however, no odds are set — instead participants bet against each other. Get your forecast right and you can win a lot of money. And if you have information nobody else has, you are almost guaranteed to do so.

Prediction markets originally came from the American DARPA — a kind of research unit funded by the US government. One of the first was called PAM, Policy Analysis Market. Its purpose was to try to forecast political developments in the Middle East, and to provide financial incentives to make the forecasts more reliable. It was a simple model: get it right and you get money.

PAM barely made it out of the starting blocks before the controversies began. In 2003 a Democratic senator called the idea “a federal gaming parlour on atrocities and terrorism” and added that the entire concept was “grotesque.” The project was shut down shortly after.

Over the past few years, a new generation of prediction markets has blossomed again. The vast majority of the activity is not political, however — it is pure sports betting. By using a legal loophole they have been able to circumvent the gambling legislation that prohibits many forms of wagering in most American states. They are now considered a type of futures contract — financial instruments — rather than a form of gambling. Not everyone agrees. The state of Nevada recently sued prediction market Kalshi in an attempt to ban them. But it is legal in the US, for the moment at least.

People betting money on a football match might sound harmless enough. But there are two major challenges with prediction markets.

The first is that it is difficult to mix two different purposes on the same market. On a stock market, there are strict rules against insider trading. Otherwise it can feel unfair that some participants in the market have access to more and better information than everyone else.

If you set aside the money, however — and simply want to arrive at as accurate a forecast about the future as possible — then insider trading is actually beneficial. You want those with unique information to share it, because that makes the forecast better. The money is only there as an incentive to bring it to light.

It goes without saying that these two motives are in conflict with each other.

Either you want to get to the truth — or you want a market with equal conditions for all.

This brings us to the second challenge. If people with secret information about, for example, military operations can make money from it — are the security of those involved not put at risk?

Already now, unusual forecasts with high betting volumes stand out. This can come to influence the actual matter at hand. If an enemy sees that an attack is expected to occur within a certain time period, it can prepare for it. The forecast was correct — but it influenced how the outcome unfolded, which can threaten the safety of those involved.

When geopolitics is mixed with gambling, a new type of problem is created. Much of the money being staked is in cryptocurrency, which makes it difficult to identify the people behind the bets. Interest has grown to such a degree that states, exchanges, and other societal institutions now follow prediction markets on a regular basis. And by betting large sums on unexpected questions, one can through this influence public opinion. It looks like an insider — but it can also be an influence operation, with prediction markets as the method.

When CNN analysed one particular player, they found that he had earned over 1 million dollars by correctly predicting how Israeli and American military operations would be conducted. The anonymous player won 93 percent of the time. The prediction market arrived at the truth. But the military was left with an enormous security problem — and a market that pays millions for valuable and confidential information to leak out.

We have transformed security policy into a kind of unregulated, anonymous stock exchange. That it would lead to problems is something everyone should have been able to foresee — even without a prediction market doing the work.

Social media is facing a crisis

March 25, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 25th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

Meta and YouTube caused a young girl to suffer anxiety and depression. An American jury has now established this in a landmark case. The verdict opens the door to an avalanche of similar cases — and could tear down the tech giants’ unique legal protection.

“Section 230 of the Communications Decency Act of 1996.”

This brief piece of American legislation, just 26 words long, is what has legally protected social media companies all these years.

Responsibility for all content uploaded — images, texts, videos — is considered under it to belong to the person who uploaded it. Services like Instagram, YouTube, and TikTok have been regarded as neutral platforms.

But that came to an end on Wednesday evening when an American jury — drawing on other laws as a basis — showed that they could be found liable for harm nonetheless.

The method is legally creative. Both Donald Trump and his predecessors have said they want to repeal “Section 230.” None of them have managed to do so, and a completely different approach was therefore needed to test the responsibility that social media companies bear for young people’s wellbeing.

What was done in this pilot case is to advance the thesis that it is the design of the services that has harmed young people — not the content. It is, for example, the recommendation algorithms and the ability to scroll endlessly that caused the harm. In this way, prosecutors were able to circumvent the deadlock that has existed when it comes to holding social media companies accountable.

The main figure in the case is a young woman now aged 20. Of the many thousands of cases making similar claims, hers was considered among the stronger, and she was therefore a good candidate to test first. What follows is therefore an avalanche of further cases in which predominantly young people will attempt to obtain redress for harm they believe they have suffered through their use of social media.

The fine of 3 million dollars — around 28 million kronor — is a rounding error for both Meta and YouTube, who are found liable. Going forward, there may be a larger class action, or a near-endless number of smaller cases with individuals making similar claims. But above all, the symbolic value of the loss is enormous. Many have drawn the comparison that this resembles the cases brought against tobacco companies in the 1990s. These proved expensive for those companies, but they also marked a turning point in public perception. Tobacco stopped being accepted in the same way after that.

It is entirely possible that the same shift in attitudes will happen here.

Parents in particular have long been worried about the impact they perceive social media to have had on their children. Research on this has been slow and insufficient, and some of the strongest evidence of harm has, ironically, come from inside companies like Meta. Internal documents and studies have shown how certain features could be addictive for young users. And yet nothing definitive has happened — beyond individual bans for young people in certain countries, such as in Australia.

After Wednesday’s verdict, these legislative proposals are likely to accelerate sharply around the world. The American ruling does not technically affect how other countries need to act, but in practice it is likely that it will do so anyway. It shifts from having been an uncertainty to a decided case in which these social media companies have been proven to have harmed a young woman. Which politicians will be able to allow this to continue unimpeded in their respective countries?

Companies like Meta, YouTube, and TikTok now face a crisis. They have numerous emerging bans for young people around the world, an enormous number of legal cases of a similar nature, and a potentially shifting opinion on how to regard these companies’ social responsibilities.

From having been a self-evident part of many young people’s daily lives, parents at least are likely to use the verdict as grounds for trying to restrict it. Some advertisers may come to want to avoid being associated with these platforms.

Earlier this week Meta also lost another case in the American state of New Mexico, for failing to protect children from abuse. Despite what are likely to be endless appeals before final verdicts are reached, the case involving the 20-year-old woman marks a clear turning point. It may be that social media has reached its peak — and is now beginning a downward journey.

Not everyone will delete TikTok from their phone immediately, but a new era is beginning. Social media can harm young people. It will now be difficult for anyone to ignore this.

Can she stop the flight to the US?

March 19, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 19th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

The EU is reaching out to the startup world with its new proposal “EU Inc.” Regulations for starting companies are to be simplified. But it misses the region’s most pressing problem.

It is the day before Christmas Eve 2025. In Berlin, a startup is trying to finalise an investment.

There is just one mandatory step left. For the process to be valid, the company’s management must sit and read all the documents aloud to a notary. The whole spectacle takes over five hours.

This is the reality in large parts of the EU. An excessively slow and bureaucratic system that complicates something other countries handle with ease. After persistent lobbying from the startup world, this understanding is beginning to sink in, even in Brussels.

The new proposal “EU Inc” is one example of this.

The European Commission hopes it will make it easier for new European companies to be started, and ideally for them to remain in the region too. Among other things, it will be possible to start a company within 48 hours, and it must not cost more than 100 euros to do so. Processes are to be digitised, and it will be possible to register a company across the whole EU through a single interface rather than going through each individual EU member state. It is to be in place by 2028.

This might not sound like a revolution to anyone who has never tried. One can go so far as to say it is strange that it does not already work this way. Given the example of the German notary, there are clearly improvements to be made. In Sweden, however, these problems are minimal. Starting and running a company digitally is very smooth. The Swedish Companies Registration Office and the Swedish Tax Agency have made otherwise quite complicated processes simple. Those who have gone furthest of all are Estonia, where you can even become a digital citizen.

While the regulatory simplifications of EU Inc are welcomed by many startups, one should not exaggerate their impact. The primary problem the EU has is not about starting companies — it is the opposite. Where do these companies end up when they eventually become large and successful? And what happens along the way?

Because the real problems start early, but long after a company’s registration.

It is about access to capital in the early stages of a company’s life. It is easiest if entrepreneurs seek money from someone in the same city, but if that is not available, one must look further afield. Despite the home continent being large, European companies often find it easier to obtain money from the other side of the Atlantic than from nearby. And if you want American venture capital, investors prefer it if you have a company registered in the US. The standard is to do so in the state of Delaware, which has the most predictable rules for all involved. Swedish Lovable is registered there, for example.

Then comes the next problem. What happens when startups become established companies.

This became clear when the EU’s legislative package the Digital Services Act (DSA) was presented and was supposed to hold the tech giants to account. Which European companies were large enough to be considered tech giants that should be regulated?

Zalando, Booking.com, and a handful of Cyprus-registered pornography sites. There are no large European tech companies of the same calibre as Meta, Google, and Amazon.

First and foremost, American companies are happy to acquire promising European companies early. Google’s AI lab DeepMind, founded in the UK, was acquired back in 2014, long before the current AI boom began. Swedish-Estonian Skype was bought by Microsoft. Finnish Wolt was bought by DoorDash. The list is long.

The second reason is that there is no common European stock exchange to list on. Spotify and Klarna are both on the New York Stock Exchange, despite having Swedish roots. The alternative for these companies might have been London, or Frankfurt if one is extremely generous. But realistically, neither was genuinely in the running.

Every individual stock exchange in the EU is too small and lacks the liquidity needed for a good investment climate suited to international companies. In the Nordic region alone there are five different stock exchanges. That is far too fragmented to function well. And so companies seek out the US instead, where they can access capital from investors around the world.

The EU simplifying regulations for startups is good news for entrepreneurs. But it sidesteps the biggest problems entirely. The startup world’s gravitational pull runs strongly westward — at every stage of the journey.

American venture capital leads to American company registrations. Which are easily acquired by large American companies. Or alternatively listed on American stock exchanges.

Until those challenges are addressed, the European startup scene is likely to remain much as it is. EU Inc makes European Commission President Ursula von der Leyen look bold and strong. But if you ask the wrong question, you get the wrong answer. And that is precisely what has happened with EU Inc.

A collapse in confidence — and it could get worse

March 16, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 16th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

Kinnevik’s CEO is dismissed on the spot, a short-seller report sends the share price tumbling — and a newly installed board member is to conduct an “impartial evaluation” of the business. The crisis at the venerable investment company continues. But why did it become acute over the weekend?

There are really only three ways to leave as CEO of a major listed company. You get a new job, you retire — or you get fired.

That is why Kinnevik’s plan in November 2025 seemed very strange. Georgi Ganev, who had been CEO of Kinnevik for eight years, was to resign. But at the same time he was to remain in post until a new CEO was appointed.

That interim period came to a definitive end over the weekend. Ganev left with immediate effect. After Kinnevik’s dismal performance over many years, no one is surprised — except perhaps by the timing. What was it that meant the plan presented four months ago suddenly had to be redrawn?

The press release Kinnevik issued on Sunday states that a newly appointed board member, Rubin Ritter, is taking over as acting CEO. He is to conduct a “thorough and impartial evaluation of our team, our culture, our ways of working and our portfolio, as well as implementing changes where necessary so that my permanent successor can get started immediately.”

Of course that is not how it will turn out — unless Rubin himself takes over as permanent CEO.

Running an investment company is not an administrative role. The job consists of setting a strategy, building a team, and making good investments. The idea that a new CEO would walk in to a fully laid table and simply do what Rubin has decided is therefore not particularly plausible. A CEO will demand to put their own stamp on the company. Shareholders should reasonably agree with this.

The talk of an “impartial evaluation” therefore smells of something else. What is it that requires evaluation?

The short-seller firm Ningi Research has a list of their own suggestions as to what it might involve. When they presented their critical view of the company last week, the already depressed share price fell 17 percent. Ningi Research pointed to, among other things, favourable transactions conducted with parties they considered to be related. Kinnevik denied the claims.

It is not in itself unusual for activists like Ningi Research to swing with a sledgehammer to achieve an effect. But that the effect would be so powerful was likely a surprise even to them. The majority of what Ningi Research wrote was already known from before. One might consider the content remarkable — but it was not new.

The share price reaction — minus 17 percent in a single day — therefore says something much bigger than a short-seller report containing old news. It says that confidence in Kinnevik’s business is at rock bottom. If it takes no more than a couple of paragraphs to erase a fifth of one of Sweden’s most venerable investment companies, that is a crisis. A major crisis.

That is why Georgi Ganev is leaving. But the above is something many people — including myself — have written several times before. It was true in November when Kinnevik announced his departure, it is true now — and it was true long before Cristina Stenbeck decided to take back the reins and make changes. Why did it suddenly become acute?

One might have expected an explanation for this from Ganev himself. No such explanation was forthcoming, however, as he made no statement in the weekend’s press release.

Finding a new CEO is now priority one, two, and three for Kinnevik’s board. And it is urgent. Having an experienced business leader like Rubin Ritter take over is a stable temporary solution. For many years he ran one of Kinnevik’s most recently successful holdings, the e-commerce company Zalando. The relationship and trust with Stenbeck have been built since then.

Being CEO of an investment company is, however, something entirely different from running a fast-growing e-commerce company — even an exceptional one like Zalando.

And above all — being CEO of Kinnevik specifically is something else entirely.

There is a heritage, a culture, and an anchoring among shareholders on the Stockholm Stock Exchange that places the company under a spotlight that its competitors do not receive. Despite those competitors often having performed substantially better than Kinnevik over many years.

From the outside, it looks as though Rubin is to ensure that the short-sellers are not right. And to clear out any irregularities. But the most important question cannot be answered until the next permanent CEO is in place: what does Kinnevik want to be?

Until the board, management, and shareholders understand and accept that answer, the crisis risks only deepening.

The threat to Lovable: their own partner

March 13, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 13th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

Billions are flowing into the Swedish AI successes Lovable and Legora. Everything seems to be pointing upward. But major risks are now emerging — and they do not come from the competition, but from a considerably closer and more sensitive direction.

In the Apple world, it is called being “sherlocked.”

The name comes from the search service “Sherlock,” whose new version Apple launched in 2002. In doing so, the privately developed search tool “Watson” — which had been created precisely as a complement to Sherlock — was rendered completely pointless.

Apple had suddenly replicated everything Watson did.

The phenomenon keeps happening. In recent days it emerged that an American company plans to spend 26 billion dollars over the next five years building new AI models. Despite the scale of the investment, it seems to have gone somewhat under the radar.

So who is this player trying to challenge giants like OpenAI and Anthropic?

It is Nvidia. The company that has supplied chips to all the world’s AI giants may now be about to compete with them.

Having to fight against one’s own partner is something many companies know well — including in Sweden.

Last year, part of Swedish Truecaller’s service was sherlocked when Apple launched a way to filter out calls from unknown numbers. Since Apple owns the hardware, they can easily see which functions are popular, and from that assess whether it is worth doing the same thing themselves.

It is a brutal asymmetry. Truecaller needs Apple to reach all iPhone users. Apple does not need Truecaller at all.

It is with these glasses on that one should read the good news from the Swedish AI companies Lovable and Legora.

Lovable reports that it has grown its revenues to 400 million dollars per year, and Legora is raising new billions in venture capital at a valuation of around 51 billion kronor. They have an enormous tailwind. Legora’s CEO Max Junestrand stated that they had not yet managed to spend the venture capital from their previous funding round, but that it was a good moment to continue building their cash reserve. Fair enough. Helping lawyers become faster and better with AI is a large market that may well require a great deal of money.

Junestrand is most likely focused primarily on beating Harvey — the American equivalent of Legora. They are even better financed and are operating on home turf in the biggest market in the US. Both are trying to be first to sell themselves to law firms as lawyers’ best AI companion. But could there be a considerably larger threat much closer to home?

Lovable is an even clearer example. They help customers build digital projects of various kinds. It is called “vibe coding.” It sounds good and is presented rather like an entirely new behaviour — and a virgin market. Lovable is at the forefront of this trend.

But it is not quite that simple. At both Legora and Lovable, it is AI models from other companies that do the really substantial work in the background. Lovable’s customers often use Anthropic’s AI model Claude, which in turn uses Nvidia’s chips. The value chain is therefore complex. And it becomes more precarious when players early in the chain decide to make a move.

The AI model Claude has a mode called Claude Cowork. There, just like with Lovable, you can build technical projects without needing to understand the code yourself. And if you want help with a legal agreement, you can go directly to Claude or ChatGPT and ask for yourself.

Claude and the Swedish AI companies are of course not identical products. But the similarities are many, for understandable reasons. At the core, it is the same service performing a large part of the work.

The challenge for AI companies that have built a large part of their business on top of others’ services is therefore to build a sufficiently good service — but one different enough from what their own suppliers might themselves be inclined to build.

And to ensure that those same suppliers do not simply switch off the functionality one depends on. Anthropic — the owner of Claude — switched off the possibility for competitor xAI to use their service.

It might sound unlikely that this could happen to Lovable and Legora — but it is not unthinkable.

As the Apple examples show, things can move fast when your best partner pivots to become your worst competitor. Right now, Anthropic and OpenAI are experiencing exactly this themselves with their own supplier Nvidia. The method is therefore well established.

What sounds like an implausible side project from an important partner can quickly become existential. And the list of companies that have disappeared in this way is long — and the last name on it has not yet been written.

The Swedish AI companies — with all their new billions — would do well to keep this in mind.

The gap reveals the stock market’s secret

March 5, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 5th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

How are tech companies really performing on the stock market? SvD’s review shows that someone who invested 100,000 kronor a year ago could today have 91,000 kronor more — or less. And it all comes down to a single question.

The Magnificent 7 is the group of tech companies that has been talked about most on the stock market in recent years. Borrowing its name from a classic western film, these seven tech giants have enjoyed a very strong run on the stock market for several years.

But the Magnificent 7 is not the only part of the tech market that has been performing well. A new SvD review shows that another category has had a substantially better return over the past year.

By comparing different segments of listed tech companies, SvD has established what has developed well — and less well — over the past year.

What emerges is a crocodile gap, where the graphs for the different segments separate so dramatically that they resemble those reptiles. Simply put, one curve points downward and the other upward.

What is the difference between these tech companies?

It is very simple. If you invested in hardware — chips and semiconductors — things have gone extraordinarily well. If you invested in software, you have instead lost a great deal of money. The Magnificent 7, which consists of a combination of hardware and software companies, sits somewhere in the middle.

For our hardware segment, we have selected the ten largest American companies that produce chips and semiconductors. These are familiar names — Nvidia, of course, but also companies like Marvell Technology, Intel, and Broadcom. Together they have risen around 68 percent over the past year.

One might think it is Nvidia pulling the entire segment upward, but that is not the case. We have equally weighted all the stocks so that size does not distort the comparison. There are plenty of individual success stories beyond Nvidia too, such as AMD (+102 percent) or Micron Technology (+355 percent). NXP Semiconductors has, however, had a tougher period, rising just under four percent over the year.

The reason for the hardware boom over the past year is AI. A stronger tailwind for a company producing chips is hard to imagine. The use of AI services drives demand for data centres, and data centres in turn must be filled with hardware.

On the software side, we have looked at the Bessemer Cloud Index, which consists of 64 companies. Over the past year it has fallen around 22 percent — nearly a fifth of its value. Companies like the HR firm Workday, the security company Okta, and the document signing service Docusign have all had weak years. And despite the company C3.ai perhaps having the most topical stock ticker on the market — “$AI” — it has fallen more than 60 percent.

Ironically, the reason software has performed so poorly is likely the same reason hardware has succeeded. It is about AI. On the Bessemer Cloud Index one finds many companies that sell services by subscription — so-called SaaS companies, which stands for “software-as-a-service.”

The fear that these companies will face compressed margins because of AI has been considerable over the year. Some have gone so far as to call it the “SaaSocalypse.” The idea is that companies can easily build their own services with the help of AI, which then leads to price pressure and lost revenues for the SaaS companies.

Finally, there is the Magnificent 7. In any other context, their performance would still look very strong. Over the year they have risen almost 30 percent on the stock market. And this from an extraordinarily high starting point. All seven — Nvidia, Apple, Alphabet (Google), Microsoft, Amazon, Meta, and Tesla — are valued at over 1,000 billion dollars. Nvidia is the largest, valued at almost 4,500 billion dollars.

In summary, someone who invested 100,000 kronor in tech stocks a year ago has experienced very different outcomes depending on their choice.

If you chose the Magnificent 7, you would have 129,000 kronor today.

If you chose the chip companies, the equivalent figure is a full 169,000 kronor.

And if you chose the worst in the tech class — software companies — you would have only 78,000 kronor left.

Stock prices were taken on 3 March and have been rounded for clarity. All companies within each segment are equally weighted, indexed at 100 one year prior.

The incident that makes the AI giants tremble

March 3, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on March 3rd, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

The Gulf states’ billions have become a financial safety net for AI giants like OpenAI, Anthropic, and Nvidia. Now the conflict in Iran is shaking the stability on which these investments rest. If the money freezes, the effects could be brutal.

On Sunday, an unusual update appeared from AWS, Amazon’s cloud services. Ordinarily, entries in the “service status” category are brief, dry notices about when a company’s services are temporarily not functioning as they should.

This time it was something else entirely.

An AWS data centre in the United Arab Emirates “was struck by an object that hit the data centre, creating sparks and fire.” The object in question was a drone from Iran.

A single fire can be resolved. But as a signal about conditions in the region — and its enormous AI investments — many are likely to be worried.

One of the most stable financiers of the AI boom is now suddenly trembling.

In the spring of 2025, Donald Trump and the Saudi Crown Prince Mohammed bin Salman stood at the front of a large group photograph. It was an investor meeting in Riyadh where the US and Saudi Arabia met to do business.

Standing in the row behind Trump was one of the main protagonists: Jensen Huang, CEO of Nvidia. He was responsible for one of the really big deals. In a first consignment, 18,000 Nvidia chips were shipped to Saudi Arabia to launch a new, enormous data centre. Hundreds of thousands more chips will follow over the next five years in a deal estimated to be worth over 180 billion kronor.

The region’s AI ambitions extend substantially further — and wider — than that, however.

The Abu Dhabi fund MGX is a co-owner of both OpenAI and xAI as well as Anthropic. Together with Blackrock, they bought the American data centre company Aligned Data Centers for around 365 billion kronor. MGX also has connections to European AI ventures, being part of a consortium that will invest at least 280 billion kronor in building data centres in France.

The corresponding funds in Qatar and Saudi Arabia have also invested hundreds of billions of kronor in total across both data centres and individual companies. Elon Musk’s xAI announced that it received money from Saudi Arabia as recently as a couple of weeks ago. The list of all AI-related commitments could be made very long indeed.

Through investments, acquisitions, and partnerships, the Gulf states have become the global AI industry’s financial safety net. The question many will now be asking is how reliable that safety net will be in the period ahead. Can one trust that the money that has been promised will materialise? What does appetite for investment look like going forward?

A large portion of the sums mentioned above have not yet been paid — they are statements of intent and plans stretching over many years. This means the risks are deferred to the future, which creates uncertainty.

Neither capital nor ambitions need to change for disorder to emerge. A temporary moment of doubt is enough to create a problem. Something as simple as project delays strikes directly at companies that may have made themselves dependent on them.

Worse still — it can take a long time to normalise. A prolonged conflict in the region would, for obvious reasons, also be able to shift focus away from long-term AI investments towards more pressing matters closer to home.

The dependence on Gulf state money appears as a significant vulnerability for the tech industry. The many data centres to be built require billions upon billions to get off the ground. If financing for these is delayed — or, in the worst case, cancelled — a cascade of consequences could quickly follow.

Supply chains break down, permits need to be reapplied for, questions of liability need to be resolved. There are many companies whose valuations — and promises to the market — depend on these projects. Were the financing to fall through, many companies could find themselves in acute financial crisis. Protracted projects delay both the construction of data centres and the AI companies seeking to purchase their capacity. It could become very expensive, very quickly.

It began with a data centre that was fired upon. But it rapidly became a wake-up call for the entire industry.

The billion-dollar windfall from the missed mega-deal

February 27, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on February 27th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

Netflix gives up on buying HBO and its parent company Warner Bros Discovery. A new major streaming giant is being created in the US instead. But despite the loss, Netflix comes out the winner.

Sometimes you win in business even when it looks like you’re losing.

Netflix is experiencing that right now.

Late on Thursday came the news that the Warner Bros Discovery board had changed its mind. In December, they agreed to sell the digital parts of their company — including HBO — to Netflix for 83 billion dollars, including debt. A good deal for all involved, many analysts thought.

One party was, however, very unhappy: Paramount Skydance, the other bidder in the process. And they made clear to Warner Bros from the start that the match was far from over.

They revised the bid structure. Changed the terms. Increased the amount of money.

In the end, it was enough for the Warner Bros board to switch sides and sell to Paramount instead. But Netflix leaves the deal as a winner — in several ways.

When making deals of this calibre, there are many things that can go wrong. All three companies are listed on the stock exchange, which means one must follow all applicable rules. Get enough unhappy shareholders and the process alone can become messy and take a long time.

After that, the deal must be approved by regulators. That too can drag on, and does not always end up as planned. The purpose is for authorities to ensure that competition in the market functions as it should. In this particular case, however, it is likely to proceed smoothly, as Paramount stands close to the Trump administration.

To give both parties some certainty, a so-called “breakup fee” is usually set. This means that if either party pulls out, or if the deal is not approved, one party must pay the other a large sum of money. It functions like an insurance policy.

As a consolation prize for Warner Bros pulling out, Netflix will therefore receive 2.8 billion dollars — around 25 billion kronor — as a breakup fee. Not bad for a few months of negotiation. And to make it even better for Netflix — as part of the negotiation, Paramount Skydance has agreed to pay the penalty out of its own pocket.

The question one might ask is why it is so important for Paramount to own Warner Bros? To such a degree that they are now paying 31 dollars per share for a company that was trading at around 12 dollars last autumn?

It is primarily about company size, and how quickly one can get to becoming a giant player.

The biggest in the streaming category is Netflix. That is the company everyone else is chasing right now. And after a market that for several years consisted of a myriad of small streaming services, the prevailing strategy is the opposite. There should be few — and large — services.

Netflix cannot be bought, and nor can Disney+, Apple TV+, or Amazon Prime Video. The list of possible acquisitions if you want to become a new giant is very short. And right at the top of it was Warner Bros Discovery. That is why this purchase became so important for David Ellison, CEO of Paramount Skydance. How else could he compete with the other massive rivals?

And so it became expensive. Paramount Skydance is a company that loses money and already has large debts following its first major acquisition — in the summer of 2024, they bought the company they subsequently took their name from: Paramount.

On the debt side, things are getting substantially worse for Paramount Skydance. They are buying Warner Bros with money they do not have. Even before this deal, they had around 124 billion kronor in debt. And now they need to cough up a further 1,000 billion kronor to pay for the new deal. All while the underlying business is bleeding.

That headache is David Ellison’s to solve going forward. Perhaps with the help of his father, Larry Ellison, the founder of the database company Oracle — and one of the world’s wealthiest people.

Netflix emerges from this as a winner in many ways. As noted, they receive a large bag of money without really having to do very much. They now have a larger competitor in Paramount Skydance, but one that has almost certainly overpaid for its new asset. The money Netflix didn’t have to spend can be invested elsewhere.

The final bonus was delivered yesterday, after the market had closed. When it emerged that Netflix had lost the deal over Warner Bros, their share price jumped more than seven percent immediately. Netflix’s market capitalisation thereby increased by 225 billion kronor on the strength of that news alone.

Sometimes, as noted, it is good to lose deals, after all.

We should talk more about how AI uses us

February 26, 2026
by Björn Jeffery
SvD Näringsliv

This analysis was first published in SvD Näringsliv, in Swedish, on February 26th, 2026. This piece was translated from Swedish by Claude. Some phrasing may differ from a human translation.

AI is not quite as artificial as many believe. A new investigation by SvD reveals how private moments are shared with office workers in Africa, whose job it is to train AI systems. Invisible to the user — but absolutely crucial to the company.

We talk a lot about how we can use AI.

Perhaps we should talk more about how AI uses us.

In an investigative report by SvD and GP, we get a glimpse into what jobs in what is called “Silicon Savannah” look like. At a subcontractor for Meta, people in Nairobi sit and watch video clips. Their job is to tell AI systems what can be seen in the footage.

It is called “annotating data,” but it can be explained more simply than that. It is when a human helps AI understand what is happening on screen, in the hope that one day it will be able to manage on its own. But before we get to that point, it needs help — drawn boxes with explanations. And it needs the raw material — the data — to get there.

In this case, the video clips come from Meta Ray-Ban — a pair of glasses equipped with a camera and microphone. With them, the company obtains a type of data that is hard to come by otherwise. If someone went around filming you on the street, you would probably ask what they were doing. You would likely speak up and ask them to stop. But when it happens through a pair of glasses, it is not as obvious what is going on.

For AI systems to get better, they use the data we produce.

There are plenty of cases where this exchange of data can be beneficial for both parties. Having one’s genetics reviewed can help identify health risks. In some cases, it can be the missing piece for understanding why one feels the way one does.

A considerably more trivial exchange occurs every time you go on Facebook or Instagram. They know who you are, and they can indirectly sell that information to advertisers who want to reach you. What you get back is an entertaining service with amusement. You have agreed that they may do this — otherwise you would not be able to use these services. But you probably do not think about it every time you use them.

While the festive photos people share of themselves on Instagram are a conscious choice, the cameras on Meta Ray-Ban face the other direction.

The report describes how workers in Nairobi are required to watch a long series of private events — sex, toilet visits, intimate conversations — as part of the job. Situations that could not reasonably have been intended to be shared with anyone, yet ended up on a screen in Kenya. Did those who were filmed even know this could happen?

It is said that you cannot walk many metres in London without being captured by one of the roughly one million surveillance cameras in the city. Signs inform people of this, but it cannot be opted out of even if one wanted to. It is the price you pay for moving around the city. The phenomenon is therefore not unfamiliar, but now it can happen anywhere. In changing rooms, at a restaurant — or in your own home.

The purpose also differs. One can certainly have opinions about the surveillance cameras that exist, but the purpose is to be able to resolve, and hopefully prevent, various types of crime. Now we are putting cameras on our faces to let an AI service tell us what we could reasonably already see for ourselves.

For the time being, artificial intelligence is not always entirely artificial, and not always particularly intelligent either. There are thousands of people sitting in offices in Africa whose job it is to tell these systems what the difference is between an armchair and a sofa. They train AI systems to become smarter.

The human component in all of this work is almost invisible to the user, but is absolutely crucial. As a result of this, a great deal of information and data about you ends up on the screen in front of individuals who watch, assess, and review what you do. They are working. But not for you — for AI.

The work was halted after SvD’s and GP’s investigation. Now 1,108 workers have been given notice of redundancy, according to multiple international media outlets.

In an office building on Mombasa Road in Nairobi, thousands of the AI revolution’s heavy labourers work. People who are supposed to teach the systems to understand what is captured in images.

They are called “annotators” and their work is invisible to the end user. The company they work for is called Sama. Until recently, its major client was the tech giant Meta, which is investing heavily in camera-equipped glasses powered by AI technology.

SvD and GP were able to reveal in February that the annotators had been exposed to a great deal of private material in the course of their work. It could involve sex, naked people, and toilet visits filmed via Meta’s glasses.

The consequences for the tech giant have been severe. Four data protection authorities on two continents have acted against Meta by demanding answers from the company or opening their own investigations.

Earlier this week it became clear that Meta is completely halting its collaboration with Sama in Kenya after investigating the claims that emerged in the investigation.

Now the consequences are falling on Sama’s employees. 1,108 people risk losing their jobs as a direct result of Meta’s decision, the company states in a written announcement.

“We are aware of the significant impact this has on the team and the local community. We are actively working to support the affected employees with care and respect,” the company writes.

The news has been reported by, among others, NTV Kenya, The Guardian, and AP. Sama states in its announcement that negotiations with Meta have been ongoing, but that the talks have been fruitless.

Kauna Malgwi, who previously worked for Sama, tells The Guardian that the layoffs show how the AI industry works.

“The power lies with the big tech companies and the risk is pushed downwards and affects the workers, often in the Global South, who have the least protection.”

Meta has stated that the company has investigated the claims in SvD’s and GP’s investigation.

“Last month we paused our work with Sama while we investigated these claims. We take them very seriously. Images and videos are private to users. People review AI content to improve product performance, for which we get clear user consent. We have also decided to end our collaboration with Sama as they do not live up to our standards.”

Meta has not, however, responded to follow-up questions about what its investigation has found. Sama has stated that the company “adheres to strict standards of data security and privacy, including GDPR and CCPA.”

After SvD’s and GP’s investigation, the Facebook owner is ending its collaboration with the annotation company Sama in Kenya. At the same time, employees speak of retaliation, insecure working conditions, and wages not paid on time.

According to multiple sources, after the work stoppage employees have been going to work without receiving any tasks. Instead, they have been sitting in front of their computers refreshing the screen roughly every eight minutes to show that they are present.

“Nobody is working on annotation in these projects anymore,” says one source.

After the investigation, employees at Sama report that conditions have worsened and that security at the company has been tightened further, and that they have attempted to track down sources.

Eric Mugendi, editor-in-chief of Kenya-based Africa Uncensored, one of East Africa’s largest investigative journalism outlets, says that the tactic of trying to identify sources is common in the country.

“After revelations like these, it is common for pressure on employees to increase, with stricter controls, and then they try to trace who has spoken to the media,” he says.

The slightest rumour can be grounds for dismissal, he adds.

“If one person does something, everyone can be punished. And if you hear something — even if it is just a rumour without concrete evidence — it can still be used against you.”

SvD and GP put questions to Sama about their view of the claims regarding worsened working conditions and the tracking of sources, and about Meta pausing its work with Sama. Representatives for Sama write in a comment that they do not comment on specific client relationships.

Sama also writes that it takes data security very seriously. “We strongly dispute several of the allegations and emphasise that we adhere to strict standards of data security and privacy, including GDPR and CCPA. We are equally committed to the wellbeing and fair treatment of our employees,” Sama writes, adding: “As a company, we do not monitor or track employees with the aim of identifying individuals who may have spoken to the media, and we reject any insinuations of retaliation.”