THE ECONOMIST: The way forward for OpenAI in 2026 as Sam Altman’s AI ambitions collide with mounting losses | USAEMALL.com

Sam Altman is like a juggler on a unicycle. Building all-knowing chatbots powered by cutting-edge artificial-intelligence models is too meek an ambition for OpenAI’s boss.

To keep his audience rapt, Mr Altman has thrown ever more balls into the air. Custom chips? Of course. E-commerce? Why not. Business consulting? Too easy. A consumer device? You betcha.

At the same time, Mr Altman must keep a hand free to hold out his cap, for the show gets more expensive by the day. Leaked figures indicate that OpenAI expects to burn through $US17 billion of cash ($25b) in 2026, up from $US9 billion in 2025, and that its losses will continue to pile up in each of the subsequent three years.

It has already raised more than $US60 billion from investors, more than any private company ever, nearly all since ChatGPT shot the previously obscure AI lab to prominence in late-2022.

Mr Altman’s hunger for cash is far from sated. In 2026 OpenAI will “almost certainly” raise another slug of capital, says a source close to the company. It is reportedly looking for as much as $US100 billion, at a valuation of perhaps $US830 billion, up from $US500 billion in its latest fundraising round in October.

Amazon is separately in talks to invest up to $US10 billion in the model-maker, which has untangled itself from an exclusive relationship with Microsoft, Amazon’s biggest rival in cloud computing. Nvidia has said it, too, may invest up to $US100 billion in OpenAI in $US10 billion increments to help it buy the chipmaker’s wares.

Although he has poured cold water on the idea at various points, Mr Altman is also rumoured to be mulling a public offering.

OpenAI’s unprecedented fundraising has fuelled unparalleled growth.

In 2023 its revenue passed $US1 billion. In 2025 it reportedly hit $US13 billion, rising to an annualised rate of $US20 billion by the year’s end. It took Google and Facebook five and six years, respectively, to pull off the same feat.

The trouble is that OpenAI’s demand for computing power — by far its biggest cost — has remained tightly coupled to revenue.

The company’s computing needs grew from 200 megawatts in 2023 to 1.9 gigawatts (GW) in 2025. It has signed letters of intent to add a further 30GW of capacity over the coming years at a cost of around $US1.4 trillion.

For now, investors are enthralled by Mr Altman’s dazzling feat. Eventually, however, he will have to prove that he can turn a profit.

Mr Altman continues to argue that OpenAI’s economics will improve as the company grows and the upfront cost of training its models dwindles by comparison.

Yet training costs keep mounting as OpenAI contends with stiff competition from rival model-makers. According to benchmarks collated by the Stanford Institute for Human-Centred Artificial Intelligence, the gap in performance among the most advanced models has narrowed significantly over the past year.

Lately the biggest threat to OpenAI has been Google, whose Gemini 3 model, launched in November, outperformed OpenAI’s GPT-5.1 on many measures. OpenAI counterpunched with GPT-5.2, but it was far from a knockout blow. Meanwhile, so-called open models, whose numerical parameters (known as “weights”) are freely available, have also been closing the gap in performance with closed models.

OpenAI cannot afford to fall behind in model-making. Already there are signs that ChatGPT is losing momentum. According to Sensor Tower, which monitors web traffic, it had 910m monthly active users as of mid-December, compared with Gemini’s 345m.

But Gemini is gaining ground. A recent study of big European countries by Deutsche Bank found that consumer subscriptions to the service “ground to a halt” in the summer and have since barely grown.

Recognising the challenge, Mr Altman instituted a temporary “code red” at the start of December, telling staff to pause other initiatives and prioritise improving ChatGPT.

More troubling still are reports that OpenAI is losing money simply by running its models, which many consumers access through the free version of ChatGPT.

In November Ed Zitron, a commentator known for his AI scepticism, published leaked Microsoft figures showing that OpenAI’s so-called inference costs exceeded its revenue in the first half of 2025.

To cut its losses, OpenAI could raise prices or curb access, but would run the risk of dragging down growth, particularly with competitors nipping at its heels.

All this helps explain OpenAI’s ever-expanding focus. Part of the strategy is to develop new routes to monetise its technology.

Although Mr Altman reportedly paused work on an initiative to integrate ads into ChatGPT as part of his “code red”, insiders say the company still has plans to do so in 2026.

Already it allows companies in America including Etsy, an online marketplace, and Walmart, a retail colossus, to sell their products through the chatbot in return for a fee.

OpenAI is also hoping to expand its revenue from enterprise customers, which tend to be stickier.

There its main competitor is Anthropic, a rival AI company whose Claude chatbot has become popular among coders in particular.

To catch up, OpenAI has built a consulting division that helps big companies deploy its technology, and has developed enterprise tools such as AgentKit, launched in October, which clients can use to automate work. Although consumers still provide the majority of OpenAI’s revenue, the share coming from businesses has been rising.

OpenAI’s strategy also seems to entail vertical integration, drawing inspiration from Google.

The search giant’s custom chips, which it began developing more than a decade ago, have become a big advantage in the AI race, costing it between a half and a tenth as much as an equivalent Nvidia chip.

The four billion users of Android devices worldwide also provide a vast distribution channel for its AI products. Over the past year OpenAI has signed a deal with Broadcom, a chip designer, to develop its own custom silicon, and hired Sir Jony Ive, the designer behind Apple’s iPhone, to develop a consumer device.

High-wire act

The difference is that OpenAI does not have the fistfuls of cash provided by Google’s search engine. Already some investors are growing wary.

The boss of one venture-capital firm notes that OpenAI’s losses are as big as the deficits of many national governments. He laments that, in fundraising discussions, “you are not even allowed to ask” about the amount of cash it is burning.

When questioned in a recent podcast interview by Brad Gerstner, one of OpenAI’s most loyal investors, how the company would fund spending commitments equivalent to roughly 100 times its revenue in 2025, Mr Altman snapped peevishly: “If you want to sell your shares, I’ll find you a buyer.”

OpenAI’s detractors decry what they see as Mr Altman’s hubris.

“This is the WeWork story on steroids,” says another Venture Capital boss who backs one of its rivals, referring to the once high-flying office-rental company that collapsed spectacularly under the weight of enormous debts and unrealistic growth projections.