2025 marked the beginning of the end for OpenAI

OpenAI is closing out 2025 in a perilous position after what was a challenging year for the AI darling.

The company has basked in stardom since bursting onto the scene in late 2022. But eventually, the limelight fades and the rock solid foundations it’s built alongside Microsoft in the intervening years have taken a few major blows this year.

Indeed, 2025 started with a big hit: DeepSeek’s dramatic entrance to the global AI market. As ITPro noted at the time, the launch of the company’s R1 large language model (LLM) rocked the US tech sector, wiping $1 trillion in stock value and prompting questions about the huge costs associated with AI training.

Adding insult to injury, within the space of a week, DeepSeek R1 overtook ChatGPT on the US App Store.

While it wasn’t quite the apocalyptic scenario some industry stakeholders suggested, the incident nonetheless marked the beginning of a year fraught with challenges – and ultimately OpenAI only had itself to blame.

The release of GPT-5 is a prime example here. The launch of the long-awaited AI model was slammed by users over slow responses and basic errors. Even the tone of the model was criticized, resulting in calls for GPT-4o to be brought back.

In the days after the launch, Sam Altman was fighting fires all over the place, taking to X to insist the company “expected some bumpiness” with the roll-out and admitting it was “a little more bumpy than we hoped for”.

It’s safe to say that “bumpy” in this instance was an understatement. The incident highlighted a deeper problem for the company: it’s scrambling to release more powerful, exciting models to stay ahead of the competition – and it’s failing.

OpenAI is now embroiled in a war of attrition with key competitors such as Google and Anthropic. The latter’s models, such as Claude Opus 4.5, have improved at a rapid pace, particularly in notable areas such as coding capabilities.

Google, however, is the real threat for the company and November gave us a glimpse into the scale of the challenge OpenAI faces competing with the tech giant. Gemini 3 blew GPT-5 out of the water, outperforming the model across a range of key benchmarks. While writing this, Google doubled down with the launch of Gemini 3 Flash, which is quicker and cheaper than Gemini 3 and performs better on some benchmarks.

Compare the current situation to what went on three years ago, and the outlook isn’t good for OpenAI. When ChatGPT launched, the announcement reportedly sent Google into a meltdown behind the scenes.

CEO Sundar Pichai initiated a “code red” alert, reallocating staff and resources in a bid to push out its own AI tooling. The tech giant has been plugging away in the intervening years and that calm, collected approach appears to be paying dividends now.

By contrast, under the direction of Altman, OpenAI has been locked in its own code red scramble in recent weeks. Reports from the Wall Street Journal in early December revealed Altman issued a call to action for developers, delaying development in some product areas to catch up with Google.

The resulting charging saw GPT-5.2 pushed out at breakneck speed, promising big performance improvements. That desperate scramble for OpenAI may have been for nothing, with MMMU-Pro benchmarks showing Gemini 3 Flash beats GPT-5.2 on multimodal reasoning with a score of 81.2% to 79.5%.

And so the arms race continues.

Rocky ground with Microsoft

OpenAI’s war of attrition is the least of its worries though. The company’s partnership with Microsoft - to the tune of several billion dollars in investment - was the fuel that propelled it to stardom in late 2022 and early 2023. But all signs point to a souring relationship with the tech giant.

Sam Altman’s ousting was the first big strain on the relationship in late 2023, and since then it appears things just haven’t been the same. Across 2025, the situation has been no different and Microsoft has been courting suitors elsewhere.

In early 2025, speculation about a big-money deal between Microsoft and Anthropic to integrate the latter’s models within Microsoft 365 Copilot service hit the headlines.

At the time, Microsoft told ITPro OpenAI “continues to be our partner on frontier models” and insisted the company was front and center in its future plans.

Fast forward to September, and Microsoft confirmed Claude will be available on the service. This marked a huge blow for OpenAI, signalling Microsoft’s latest shift away from exclusivity.

On one front, OpenAI is being pushed to breaking point by Google on model development and release timelines. On the other, its longtime patron appears keen on distancing itself from the company.

Altman bets big on future revenue

2025 was the year OpenAI became synonymous with the idea of a looming “AI bubble”, largely fueled by its increasingly erratic and mind-boggling spending plans.

It’s no secret that the company needs a monumental amount of cash to keep running, and questions have been raised repeatedly over its ability to basically stay afloat. This year alone the company has lined up more than $1 trillion in spending, signing megadeals with Oracle, Nvidia, AMD, Broadcom, and most recently with Amazon Web Services (AWS).

So how does it plan to pay for all this? Your guess is as good as mine. According to the Financial Times, roughly 70% of the company’s $13bn in annual recurring revenue comes from the consumer version of ChatGPT.

The first objective here appears to be a significant increase in the number of paying subscribers. At present, that’s just 5% of the 800 million users globally. Setting this goal is one thing, actually getting them to fork out is another and seems wholly unrealistic.

There’s a sense that OpenAI is handcuffing itself to so many major industry players that if the worst comes to worse, it’ll either drag everyone into the abyss and brick the economy, or force said providers to take drastic measures to keep it in the game.

Reports from Fortune in October this year noted that the lion’s share of US GDP growth in the first half of 2025 was fueled by investment in data centers infrastructure. We’re approaching a situation where if OpenAI goes kaput and drags everyone down with it, economic turmoil will follow.

OpenAI has no clear direction

The clearest sign of OpenAI’s challenge across 2025 lies in the fact the company clearly doesn’t know exactly what it wants to be.

In May the company acquired Jony Ive’s hardware firm, io Products, and speculation over the launch of a “palm-sized personal assistant” next year has been growing, per reports from the Financial Times.

This was the moment the penny dropped for me in 2025: OpenAI has no clear direction and is struggling to find its identity. It’s scrambling to appeal to consumers and enterprises alike with a myriad of paid tiers, grappling compute costs and burning through cash, and has even suggested building a new social network.

OpenAI has too many fingers in too many pies, and it’s rapidly running out of spare digits to stick elsewhere.