Why the ChatGPT inventor is suddenly disappointing expectations

• OpenAI is facing strategic challenges. The new ChatGPT model is disappointing and the important partnership with Microsoft is shaky. In addition, a number of top executives have left the start-up.

• However, the company is not only struggling with itself, competition in the AI sector is also growing.

• ChatGPT still dominates the AI market.Es gibt zunehmend allerdings Alternativen, die gezielte Vorteile bei der Suche, Bildgenerierung oder Programmierung bieten.

OpenAI needs space. The ChatGPT developer has moved out of its old headquarters in San Francisco's trendy Mission neighbourhood, the "Pioneer Building" with its beamed ceilings and wooden windows. The future is modern and spacious: OpenAI has leased 46,000 square metres from the ride-hailing company Uber on 3rd Street. The brief to the architects: as few screens as possible, as much space for thinking as possible. Now the next industrial revolution is maturing in San Francisco's glass palace - or its most expensive mistake.

OpenAI still had around 120 employees in 2020. By mid-2025, the number was a good 3,000, and that is by no means the end of the story. OpenAI recently rented 29,000 square metres in the neighbourhood and is looking for more offices.

The rapid growth is representative of the development of the once public welfare-oriented research lab into the most important start-up for artificial intelligence (AI). OpenAI triggered the AI hype with the launch of the ChatGPT text robot at the end of 2022 - and is still driving it today.

Chip supplier Nvidia and OpenAI partner Microsoft alone recently had a weighting of around 15 per cent in the broad US share index S&P 500: a record. And if the analysts at Morgan Stanley have their way, AI could create an annual economic added value of around USD 920 billion for the S&P 500 companies, more than a quarter of their annual pre-tax profits.

When Nvidia presented its figures on Wednesday, AI experts held their breath. And indeed, the Group reported growth of 56 per cent for the past quarter compared to the same period last year. However, this led to a certain amount of disillusionment on the markets, as more had been expected.

OpenAI boss Sam Altman is regarded as the mastermind behind this technological revolution.He raved to Handelsblatt about the exponential growth of AI performance in 2023"If artificial intelligence enables us to live in abundance, then this prosperity should belong to all of us." Altman emphasised the opportunities, not the downsides of the technology: job losses thanks to AI, divided societies - and data centres that are as power-hungry as entire countries.

And its top manager Brad Lightcap now emphasises these ambitions in an interview with Handelsblatt: "Very powerful, very reliable and very secure systems" are the goal, and the company is also working on super intelligence (AGI): "We have a very healthy and aggressive roadmap for the future."

OpenAI's valuation alone is about to rise to 500 billion dollars, which would make the start-up the most expensive non-listed company in the world. And if OpenAI achieves its declared goal of building a superintelligence, the sky is the limit anyway.

The rise of OpenAI is also the beginning of a new era. Its AI models are shaping software, markets and political debates. But behind the façade of glass and vision, cracks are appearing - in the technology, in the strategy, in the leadership. The presentation of the long-awaited GPT-5 model was disappointing. The close partnership with tech giant Microsoft, which once catalysed the rise of OpenAI, is increasingly suffering from conflicting goals. And on top of that, Altman has also lost several of its key executives.

A new Head of Product should now finally ensure that the company is in the black. And Altman himself is suddenly taking a more critical stance, even talking about a "bubble" in AI. Which makes the industry shudder. Where is OpenAI, the centre of gravity of the AI world and the self-proclaimed pacemaker of a new age, heading? In recent weeks, Handelsblatt has spoken to insiders, customers and analysts. It became clear that OpenAI is facing a hot autumn - and perhaps the most decisive months in the history of the company, if not the entire industry. But first things first.

1st GPT-5: How the top model fluffed its launch

Kieran Healy is not an AI researcher, but a sociology professor at the renowned Duke University. Nevertheless, he loves to explore the advantages of the new technology - and was all the more excited about the launch of GPT-5.

At the product launch in mid-August, Altman promised that GPT-5 marked the beginning of a new era. The model would enable anyone to have an AI with the skills of an "entire team of PhD students at their fingertips". Healy wanted to test this. And failed because of the word blueberry.

He was "not particularly impressed", the professor wrote on the Bluesky platform. He had commissioned GPT-5 to count the Bs in blueberries. The AI's repeated result: three. And this is just one of the many bizarre errors in the model that the internet has been making fun of since the beginning of August.

The law of scaling has applied for years: more data, more computing power, larger models - and AI gets better. Altman himself often spoke of the exponential growth of capabilities. The result was a global bet: the expansion of AI data centres as one of the largest infrastructure projects of our time. But GPT-5 calls this into question: despite billions in investment, the big breakthrough has yet to materialise.

Apple researchers speak of the "Illusion of Thinking" in a much-noticed paper": AI models solve simple tasks surprisingly well, but collapse when faced with more complex problems. More parameters or longer computing time do not help. This is confirmed,what critics like Gary Marcus have long fearedThe technology is inherently unreliable.

After the Chinese start-up Deepseek sowed doubt at the beginning of the year when it presented a powerful model with significantly less computing power, the entire industry could be in for a rude awakening with GPT-5: Is the era of almost limitless growth over? Recently, even the high priest of the AI world himself has issued a warning.

At a dinner with journalists in San Francisco a few days ago, the OpenAI boss answered the question of whether investors were collectively overvaluing the AI sector: "Yes." The current enthusiasm is comparable to the dotcom boom at the turn of the millennium,when many investors burnt themselves out on internet start-ups, even though the underlying technology was actually changing the world.

"In every bubble there is a spark of truth - and it is precisely this that often leads smart people to exuberance," said Altmanaccording to the portal The Verge. There are sky-high valuations for start-ups that have little more than a pitch deck; someone will "fall flat on their face". Are investors generally "over-excited" when it comes to AI? "I think so." But technology leaders like OpenAI should still win: "Is AI the most important thing to happen in a long time? I also think yes."

Now theonly the customers and partners still believe. But its most important partner, Microsoft, is going astray.

2 Microsoft: Why the most important partnership is wobbling

Satya Nadella stands on the big stage in Seattle and praises it to the skies. According to the Microsoft boss, the aim is to become "truly multimodal", i.e. to offer more than just OpenAI's models. He therefore had a "fantastic" special guest with him who had a "deep vision" of what AI should be: Elon Musk. Him of all people.

For Altman,also joined Microsoft's developer conference in Maythat must hurt. Musk, who co-founded OpenAI in 2015 and chose the old headquarters in the Pioneer Building, has become Altman's nemesis. For months now, Musk has been covering OpenAI with lawsuits and threats,Altman insults Altman on his platform X as a "swindler" and "fraudster". At the beginning of 2025, according to court documents, Musk even tried to get Meta boss Mark Zuckerberg fora 97 billion hostile takeover bid for OpenAI to win. And now Nadella is rolling out the red carpet for him.

Altman can hardly complain publicly. One thing is clear: without Microsoft, OpenAI would probably no longer exist. Since 2019, Microsoft has invested more than 13 billion dollars in OpenAI under Nadella, primarily in the form of computing time in its Azure server centres. To this day, ChatGPT is operated exclusively on Azure hardware, and Microsoft is incorporating OpenAI's models into more and more products.

So far, so successful. If it weren't for the growing cracks in perhaps the most momentous tech partnership of recent years. Outwardly, the joint future is conjured up. We are facing "fantastic opportunities", says Altman in Seattle, such as autonomous AI agents. "Thank you for the partnership," purrs Nadella. But in fact, both sides are going their separate ways.

Ultimately, the models working in the background of the agents are interchangeable, Ray Smith, Head of AI Agents at Microsoft, told Handelsblatt in Seattle. The technology will soon no longer play a major role and will be optimised depending on the use case. "We will talk less about the latest OpenAI and Anthropic models in the future," said Smith. "The focus will shift to which platform you create the agents on." And in his eyes, that means Azure, of course.

A clever move, believes Rowan Curran, analyst at consulting firm Forrester: "Microsoft has skilfully positioned its Azure platform as a central point of contact for AI services. While OpenAI is not allowed to distribute its models via other cloud providers due to contractual agreements, Microsoft can easily integrate models from different providers into its ecosystem." This results in "model gardens", which are particularly interesting for corporate customers, who can choose the best model for each use case. "The relationship with OpenAI is no longer nearly as critical for Microsoft as it was a few years ago," says Curran.

This is a problem for OpenAI. And the task of breaking free from Microsoft's grip is anything but trivial.

3. solo dance: How OpenAI wants to stand on its own two feet

OpenAI's internal strategy paper is classified as "highly confidential". In the document entitled "ChatGPT: H1 2025 Strategy", the start-up outlines where it sees its own future. It provides a rare insight into the inner workings of the secretive organisation - and reveals its ambitions.

One thing is clear: OpenAI wants to be anything but an appendage of Microsoft. Instead, the team wants to develop a super assistant, "a system that knows users, understands their needs and supports them in almost every task." Such a product would quickly become indispensable, but OpenAI is facing growing competition: "No advantage lasts forever," admits the paper, whichwas disclosed by the US Department of Justice in the course of antitrust proceedings against Google.

It is not known whether Brad Lightcap also contributed to the paper. But it is not impossible: the 34-year-old Chief Operating Officer (COO) of OpenAI is considered one of Altman's most important confidants. Like the other top managers, he is overshadowed by the CEO. But according to Fortune magazine, which annually selects the "25 most powerful up-and-coming executives", the world will soon know his name.

Lightcap is an exception in the AI sector: the Duke University graduate studied economics and history and worked as an analyst at the major bank JP Morgan. Lightcap worked as an investor at venture capitalist Y-Combinator, which Altman managed, before joining OpenAI in 2018. Since March, he has been responsible for OpenAI's business operations, including the partnership with Microsoft - and its further development.

The collaboration is quite normal, "a marriage with ups and downs", says Lightcap in an interview with Handelsblatt. The relationship was designed for evolution "right from the start": "We are a very fast-growing company. Our needs change frequently."

As an "important partner", Microsoft "naturally wants to ensure that it has the latest technologies". However, this fits in with its own mission, explains Lightcap. The division of labour is clearly defined: Microsoft uses OpenAI technology for its own products such as Github Copilot and offers OpenAI models via Azure interfaces. OpenAI, on the other hand, operates its own services such as ChatGPT and also makes its models available via interfaces. "It's basically the same basic technology, but two different offerings."

There will be no departure from Microsoft; in future, the company will even increase its expenditure on Azure. But yes, says Lightcap, they are also expanding their partner base. "At some point, OpenAI needs to be able to build infrastructure on a global scale to support demand. This scale could exceed what Microsoft can do as a single company."

In fact, OpenAI has long been looking for alternatives. The largest cloud contract in history was announced in July: OpenAI willPay around 30 billion dollars per year to Oracle from 2028for data centres full of high-performance chips. The two companies announced their "Stargate" project back in January,the establishment of a new company to invest 500 billion in AI infrastructure.

OpenAI has also signed contracts with Coreweave and Google Cloud to make itself less dependent on Microsoft's computing capacities. The start-up's first European large-scale data centre is scheduled to go online in Norway in 2026.

According to Lightcap, the active search for new partners is not a problem: "Our contract with Microsoft stipulates that we can scale beyond Azure if we need additional capacity." Both companies are now competing for the same customers - which is also not a problem, as they are "two completely independent companies" with no mutual influence on customer relationships.

OpenAI's strength lies in "working with fewer companies, but in a more practice-orientated way." The start-up focusses on technically demanding use cases and has acquired customers in Germany such as Zalando, Deutsche Bank and the Sparkasse Group. Germany is one of the most important markets,a new office in Munich was recently opened.

So everything on track? Lightcap is aware of the challenges: OpenAI is now one of theaccording to own information more than 700 million active users per week. However, the vast majority of them use the free offer, which means they pay more for the time being. Nevertheless, consumer subscriptions are expected to account for around 75 per cent of revenue. According to internal calculations cited by the New York Times and others, OpenAI posted a loss of around five billion dollars for 2024 - on a turnover of just 3.7 billion dollars.

Even more alarming is the start-up's cash consumption ("burn rate"): OpenAI spent around 2.25 dollars to generate one dollar in revenue. The training and operation (inference) of the models alone consume billions. Analysts at the major bank JP Morgan estimate that the start-up could burn through around 46 billion dollars in cash over the next four years to pay for computing power, inference costs and salaries.

However, OpenAI cannot keep all of its revenue. Microsoft has demanded a share in return for its financial injections, which has a negative impact on profitability. According tothe portal The Information Microsoft did not confirm this when asked, but referred to a joint statement that emphasised the "long-term, productive partnership that has produced amazing AI tools for everyone". They are "optimistic" for the coming years and are in "ongoing discussions".

German investor Frank Thelen says: "The struggle between OpenAI and Microsoft is a poker game for control, intellectual property and money." Both sides are still interested in a partnership, but the future conditions are unclear. And Microsoft could try to marginalise OpenAI "by setting up its own alternatives."

One thing is clear: OpenAI wants to renegotiate the terms of the non-public contract - the start-up had promised this to the investors in the latest financing round. However, Microsoft is refusing, as reported by the Wall Street Journal and Financial Times. A key sticking point: OpenAI's intellectual property, which Microsoft has access to until 2030. The outcome of the negotiations: open.

4. personnel: loss of top employees

OpenAI had positioned itself as an elite project: founded by pioneers, led by researchers, driven by a common mission. Altman was and is the central figure. But only one of several. The majority of the old top managers have left the company. Co-founder and chief developer Ilya Sutskever, former head of research John Schulman, AI architect Mira Murati and leading figures such as Andrej Karpathy and Barret Zoph are no longer with OpenAI.

The exodus hit OpenAI to the core: hardly any other tech company was so tailored to a small, highly specialised management team - and to the idea that only such a group could be responsible for the development of artificial superintelligence. Almost all of the eleven founding members have left the company. Many, such as Ilya Sutskever and Mira Murati, founded direct competitors.

At least OpenAI has hired reinforcements for the numerous construction sites: Fidji Simo has been leading the business initiatives as "CEO of Applications" since August.

The former head of the delivery service Instacart is to transform OpenAI from a chaotic start-up into a structured tech company with a clear profit perspective. As OpenAI confirmed to Handelsblatt, Lightcap will also report to her in future. However, the Altman confidant will remain the link to investors and customers. And Altman himself? He will focus more on research, infrastructure and long-term projects - so it was all the more important for the frontman to emphasise that he will remain CEO and be responsible for all business areas.

The new management has to settle in - and is already facing its greatest feat of strength, also in preparation for an IPO: the planned transformation into a Public Benefit Corporation (PBC). This model is intended to combine OpenAI's public benefit mission with opening up to the capital market.

Various scenarios were played out. The plan now is for the non-profit organisation to become a large and controlling shareholder in the PBC. The profit-oriented arm of the organisation, in whichaccording to Altman in future "anyone can hold shares". Competitor Anthropic is also organised as a PBC.

A look back to autumn 2023 shows the explosive nature of the project: "The weekend", as some employees call it, was OpenAI's biggest crisis to date. Practically everyone who was involved at the time can remember exactly how it happened: the near implosion. "It was the craziest weekend of my professional life," says one OpenAI employee.

In November 2023, the Board of Directors surprisingly fired CEO Altman - on the grounds that he had not been "consistently honest" in his communication.Insiders told the Handelsblatt that there was a great deal of mistrust, also triggered by Altman's side businesses.

But the palace revolution only lasted five days. 700 of the approximately 770 employees threatened to resign, and Altman eventually returned. A central role was also played here: Lightcap, who called the most important customers during these days (no one cancelled).

Even back then, the background was all about turning OpenAI into a profit organisation. OpenAI was founded as an open research laboratory for AI, whose developments were intended to benefit "all of humanity". To this day, OpenAI is controlled by a foundation-like, non-profit organisation,whose board of directors has no investorsthe profit-oriented arm founded in 2019.

But time is working against OpenAI when it comes to reorganisation. In the last round of financinga deadline of 31 December 2025 has been agreedOtherwise, an investor group led by the Japanese technology group Softbank could halve the 40 billion dollars it has pledged. The signalling effect would be even more serious: the valuation of half a trillion dollars targeted as part of an employee share sale - around 200 billion dollars more than in March - is unlikely to be maintained if hopes of an IPO fizzle out.

At the same time, political pressure is growing. As recently as April, a group led by Nobel Prize winner Geoffrey Hintonin an open letter to the attorneys general of California and Delaware in favour of preventing the conversion. This "would remove essential safeguards and thereby transfer control of possibly the most powerful technology ever created" to a company committed purely to shareholder profit.

OpenAI has responded to this andaccording to the magazine "Politico" political heavyweights, including Bill Clinton's ex-spin doctor Chris Lehane and Kamala Harris confidante Debbie Mesloh.

Internally, it is emphasised that the restructuring should not jeopardise the common good. In any case, Lightcap is optimistic that the reorganisation will succeed. "The idea is to bring our non-profit and profit units closer together," he explains. The focus on the common good will remain strong and the non-profit organisation will become "one of the largest and best-equipped philanthropic institutions in the world." At the same time, the organisation is trying to "open up all possible options" without being "restricted by its own structure".

Whether this balance succeeds will be decisive for the future of OpenAI - and whether Simo and Lightcap can fulfil their mandates as architects of the transformation. If it weren't for one last challenge: the dear competition.

5. competition: who can stop OpenAI

The cast-iron flywheel in the Y-Combinator headquarters stands still. The new headquarters of the influential venture capital investorin the Dogpatch neighbourhood is spread across a series of old factory buildings on San Francisco's harbour, just like a start-up. Brick, wood and disused machine shops from the turn of the century form the picturesque backdrop for the technological achievements of tomorrow, or so it is hoped: Y-Combinator has recently instructed all start-ups in which it has invested to locate within a mile radius of the new headquarters.

Marc Mengler also works here. Mengler, who lives in San Francisco and Berlin, has made it, you might think. He sold his former start-up, Understand.ai, for a sum in the double-digit millions. Mengler no longer has to work, but the AI world won't let him go. So he founded a new start-up, Octomind, whose AI finds errors in website code. He knows how tough the battle for the best talent is - and who comes out on top.

"Those who already have a name as an AI researcher often go to Anthropic," says Mengler, the AI start-up of former OpenAI manager Dario Amodei, which emphasises responsible AI development in particular. Those who are young and still need to make contacts, on the other hand, often go to OpenAI, which is active in more and more sectors. Meta and other companies are also vying for the best talent. The result of the battle for brains: "Salaries have reached absurd levels."

As recently as June, Altman explained that Meta had attempted to entice OpenAI employees away with starting bonuses of up to 100 million dollars and even higher annual salaries. Meta's head of technology Andrew Bosworth explained that OpenAI had made corresponding counter-offers.

Nevertheless, Shengjia Zhao, ChatGPT co-developer and former chief scientist at OpenAI, moved to Meta, which caused internal unrest. In August, Julia Villagra, who had only been promoted to Head of Human Resources at OpenAI in March, had to leave.

The competitive pressure for OpenAI is therefore greater than ever. And all the money for personnel and hardware has to be earned first - which is unlikely to get any easier. After all, the competitive landscape has changed dramatically: OpenAI once dominated the AI market with its text robot ChatGPT. However, there is now fierce competition between several players, which is driving down prices.

In addition to Anthropics Claude, free services from Meta and Google also compete with ChatGPT. In China, Alibaba, Baidu and Deepseek are among those positioning themselves. Altman publicly admitted that OpenAI is losing money with numerous subscriptions because "people are using it much more than expected".

Some analysts expect OpenAI to introduce advertising by 2026 to monetise its 700 million users, although this could jeopardise response quality and user trust. OpenAI itself predicts for 2029according to media reports a turnover of 125 billion dollars through the monetisation of free users and new business areas such as AI chips and robotics.

However, these ambitious targets require maintaining an unprecedented annual growth rate of over 90 per cent while at the same time drastically improving returns - a combination that few companies have achieved to date.

Altman's latest idea: OpenAI is to be geared towards a consumer focusfrom the AI device with ex-Apple designer Jony Ive through to partnerships with industry leaders. This makes it clear how closely business realisation (Simo) and visionary strategy (Altman) must now interact in order for OpenAI to hold its own against the competition. Is this the Apple or Google of the AI age?

6. future: what could remain of Altman's visions

The top manager of a German car manufacturer is sceptical: "Why should I go to OpenAI when I can get everything directly from Microsoft? Especially when we have to make savings?" The news surrounding OpenAI's transformation in particular has caused unease among potential customers.

Start-ups that rely on OpenAI's technology are also weighing up the options. Venta AI, for example, sees itself as a digital sales representative: the software identifies target customers, researches contact persons and creates personalised message suggestions - the technical basis is also provided by OpenAI models.

"We try to spread the risk by connecting several model providers at the same time," says co-founder Lucas Spreiter. "If OpenAI changes direction or we realise that another model is more suitable, we can switch relatively easily." Switching from OpenAI to Google or Anthropic costs the start-up around a week.

For compliance reasons, OpenAI's models are used exclusively via Microsoft Azure, says Spreiter: "Azure hosts the models in Europe." Although OpenAI itself now also offers to host the data in Europe, the start-up has been "less co-operative" with queries so far, says Spreiter. The bigger question, however, is: "To what extent will OpenAI itself compete with our business model?" Sales and marketing solutions in particular are among the first use cases that OpenAI has in mind when transforming its technology into products.

For research, the situation is simpler. The Max Planck Institute in Tübingen has a partnership with OpenAI and uses the models in quantum physics and basic research. AI researcher Wieland Brendel uses OpenAI primarily for research, "in this area, the models are currently even better than those from Anthropic". But: "If in doubt, we can quickly switch models."

Forrester analyst Curran sees OpenAI's growth targets as being jeopardised in the face of increasing competition. "What looked like a duel between OpenAI and a few competitors two years ago has developed into a crowded market." There are enough alternatives. And corporate customers "want to control their risk", says Curran. The choice of provider is also "not just about the AI models, but about all the tooling around it", i.e. databases, monitoring tools, security issues and interfaces: "Areas in which established cloud providers naturally have an advantage."

And Altman's promise to one day develop the artificial superintelligence, the AGI? Curran waves it off. "It's not important." For corporate customers, it's not the vague prospect of a super-intelligent future that counts, but the concrete economic benefits today.

Raising this requires one thing above all: money. So far, this has not been a problem, as OpenAI is well financed. However, critics fear that the cost structure could lead to a vicious circle in which further growth actually increases the relative loss - due to the energy and chip-hungry AI technology.

The optimists, on the other hand, are unperturbed. According to JP Morgan analysts, OpenAI generated around ten billion dollars in sales by July, representing an 82 per cent increase in sales in the first half of 2025. "By the end of the year, the projected annual turnover should reach 20 billion," says Earlybird investor Retterath. That is absolutely realistic.

He rejects the accusation of overvaluation: the 500 billion dollars are understandable, says Retterath. "This corresponds to a sales multiple of around 20 over the next twelve months and is therefore even lower than large listed tech companies such as Palantir, Figma or Cloudflare."

According to most observers, one thing is clear in autumn 2025: OpenAI is at a crossroads. The company that triggered the AI hype must prove that it is more than just another highly valued model provider in an increasingly crowded market.

One person is unwaveringly optimistic: CEO Altman. "OpenAI is many things today, but first and foremost we are a superintelligence research company. We still have a lot of work to do, but most of the path is now illuminated and the dark areas are quickly disappearing," he wrote two months agoon his private blog. At OpenAI, they are keen to follow his lead. And hope that the boss doesn't go astray in the semi-darkness.

Because he still has enough side businesses. World, for example,a verification system for humanity. Or Merge Labs, a brain chip start-up that Altmanaccording to media reports and invest in the OpenAI.

Sam Altman's corporate empire needs more space for the time being. And what will become of its crown, OpenAI, is likely to be the talk of the town: Promise or bubble.