[AI Bubble] SoftBank Deepens OpenAI Bet With Another $30 Billion, Profitability Risks Loom

SoftBank to Inject Up to $30 Billion More Into OpenAI
SoftBank Keeps Betting Aggressively, Selling Stakes Including Nvidia Holdings
OpenAI Struggles to Prove Profitability, With an Ad-Supported Plan Still an Open Question

SoftBank Group has decided to commit a large new tranche of funding to OpenAI, the developer of ChatGPT. After selling off substantial assets last year to deploy tens of billions of dollars, SoftBank now appears to be reshaping its investment portfolio around OpenAI as its central bet. Market watchers, however, are raising concerns that the wager may be overly risky, given OpenAI’s ongoing struggle to establish a sustainable path to profitability.

SoftBank Bets on OpenAI Again

On the 28th (local time), Bloomberg and The Wall Street Journal (WSJ), citing multiple sources familiar with the matter, reported that SoftBank is considering an additional investment of up to $30 billion in OpenAI. The move would come after SoftBank wrapped up a large-scale investment worth tens of billions of dollars last year, and would mark another push for fresh funding early this year. The talks are said to be at an early stage, and the investment amount could still change.

SoftBank previously pledged in February last year to invest $40 billion in OpenAI, valuing the company at $260 billion. In April, it made a direct investment of $7.5 billion, and raised an additional $11 billion with co-investors. By the end of last year, it injected a further $22.5 billion, bringing its total investment to $41 billion. Through those deals, SoftBank secured an 11% stake, becoming OpenAI’s third-largest shareholder after Microsoft and the OpenAI foundation, which hold 27% and 26%, respectively.

If the new investment goes ahead, OpenAI is expected to emerge as one of SoftBank’s largest holdings, and OpenAI’s valuation is estimated to climb to as high as $830 billion. Some in the market have also suggested that if OpenAI pursues an IPO, its valuation could reach $1 trillion.

SoftBank’s Funding Playbook

SoftBank’s aggressive push into OpenAI is being viewed as part of a broader capital reallocation aimed at securing leadership in the AI ecosystem. To raise funds for OpenAI, SoftBank previously sold its entire Nvidia stake worth $5.8 billion and also disposed of holdings in T-Mobile US worth about $4.8 billion. It has sharply slowed the pace of new investments through its Vision Fund, and required Masayoshi Son, chairman of SoftBank Group, to personally approve any investment of $50 million or more.

SoftBank is also pursuing an IPO of PayPay, its payments services arm, to raise additional capital. The market expects the listing to help SoftBank raise roughly $20 billion in the first quarter. Another key funding source is margin loans backed by its stake in Arm Holdings. SoftBank has recently expanded its margin-loan capacity and now has $11.5 billion in unused borrowing headroom. With Arm shares up more than threefold since its IPO, the value of that collateral has risen sharply.

Some observers, however, argue that SoftBank’s OpenAI-heavy strategy comes with substantial risk. Concerns over an “AI bubble” have been spreading, and critics say SoftBank is concentrating its portfolio too heavily in a single AI company. While investment in AI infrastructure—such as data centers, power supply, and semiconductors—has continued to accelerate, the revenue models needed to justify that spending remain insufficiently proven.

Costs Outpacing Revenue, Profitability Under Pressure

OpenAI itself is also struggling to secure profitability. To be sure, the company’s topline has shown clear growth in recent years. On the 18th, OpenAI CFO Sarah Friar wrote on the company blog that annual recurring revenue reached $2 billion in 2023, $6 billion in 2024, and is expected to exceed $20 billion in 2025, marking threefold growth each year and a tenfold increase overall. She described the pace of expansion as unprecedented.

The problem is that costs are rising even faster than revenue. According to OpenAI’s estimated 2025 profit-and-loss figures compiled by HSBC’s U.S. software and services team, the company’s projected operating loss last year reached $17.72 billion. The burden has been driven higher by a surge in mega-scale computing contracts. In 2025, OpenAI signed a $250 billion cloud computing agreement with Microsoft and a separate $38 billion deal with Amazon.

Competitive pressures are also mounting. OpenAI’s consumer paid subscription services, which account for 60–70% of revenue, have recently come under pressure from Google. Users have begun to migrate following Google’s launch of Gemini 3 and Nano Banana Pro last November. According to web analytics firm Similarweb, ChatGPT’s share of generative AI traffic fell from 87% early last year to 64% in January, while Gemini’s share jumped from 5% to 21% over the same period. In the enterprise AI market, OpenAI holds a 27% share, trailing Anthropic, which leads with 40%.

Against this backdrop, OpenAI launched a lower-priced, ad-supported plan called ChatGPT Go globally on the 17th in an effort to generate additional revenue. The strategy is to lower the barrier to entry for subscriptions, attract new users, and tap advertising income. According to a recent report by The Information, ChatGPT’s ad pricing has been set at about $60 per 1,000 impressions, on par with premium TV and streaming ads such as live NFL broadcasts and well above Meta’s average CPM of $20. Some in the industry believe advertisers may accept the high rates given ChatGPT’s massive user base, but many are likely to remain cautious until ad effectiveness is proven.

OpenAI is also directly linking advertising to commerce as part of its monetization push. A recent example is allowing U.S.-based Shopify merchants to sell products directly within ChatGPT in exchange for a 4% fee. The move goes beyond clicks and impressions to establish a revenue model tied to actual sales. Still, shopping within an AI chatbot remains far from a mainstream consumer behavior, leaving the extent of real purchase conversion an open question.