SoftBank Group has reopened discussions with a consortium of lenders for a $10 billion loan backed by its stake in OpenAI, after earlier attempts to secure the financing stalled over concerns about the difficulty of valuing private companies, according to two people familiar with the matter cited by Reuters.

To address lender hesitation, the Japanese technology investor is now offering to guarantee repayment of the loan, giving banks recourse to SoftBank itself if the OpenAI shares pledged as collateral lose value, the people said. The lending consortium is expected to include Goldman Sachs, JPMorgan Chase, and Mizuho Financial Group.

The funding forms part of SoftBank’s broader efforts to finance its ambitious artificial intelligence investment strategy. The financing, structured as a margin loan, functions like a line of credit against the value of the pledged shares.

Initially, SoftBank had sought a loan backed solely by its OpenAI stake, but banks pushed back because they would have had no claim on SoftBank beyond the collateral if the shares declined in value. Under that earlier structure, SoftBank would not have been obligated to repay the debt if the collateral proved insufficient.

The revived talks highlight lenders’ increased caution toward loans backed by stakes in privately held companies, whose valuations are harder to assess and whose shares are more difficult to sell than publicly traded stock. Reuters was unable to determine whether lenders have specific concerns about OpenAI’s valuation. Valuations of major AI companies such as OpenAI and Anthropic have ballooned in recent years amid intensifying competition for leadership in the field.

Financing an AI Bet of Historic Scale

SoftBank has become one of the world’s biggest backers of OpenAI under founder Masayoshi Son’s push to position the Japanese conglomerate as a dominant investor in artificial intelligence. It has committed more than $60 billion to OpenAI and related AI infrastructure projects, including the Stargate data center venture announced alongside OpenAI and Oracle last year.

The company has relied heavily on debt and asset-backed financing to fund these investments. In recent months, SoftBank has explored several financing options tied to its investment portfolio. Last year, it sought to raise a $5 billion margin loan backed by shares in chip designer Arm Holdings, whose stock has surged amid investor enthusiasm for AI. Unlike the OpenAI financing, the Arm loan was backed by shares in a publicly traded company, making the collateral easier for lenders to value and liquidate if necessary.

SoftBank had previously sought to raise at least $10 billion through a margin loan backed by its OpenAI stake before reducing the target to about $6 billion after encountering hesitation from lenders, Bloomberg News reported earlier.

OpenAI confidentially filed for a U.S. initial public offering in June, which could ultimately make SoftBank’s stake easier for lenders to value and eventually liquidate. The company also faces a March 2027 deadline to repay a $40 billion bridge loan that helped finance its OpenAI investment. SoftBank has said that borrowing would likely be repaid “through the utilization of existing assets and other financing measures.”

Son has accelerated the company’s spending on AI this year, making investments spanning data centers, semiconductors, and robotics as he seeks to position SoftBank at the center of the industry’s rapid expansion.

Lenders’ Caution Grows

The difficulty in securing the loan underscores a broader trend in financing, where lenders are becoming more selective when it comes to private company collateral. Valuations of private tech firms can be opaque and volatile, and liquidity is limited compared to public markets. Against that backdrop, analysts believe that SoftBank is effectively reducing the risk for lenders by offering a personal guarantee, potentially making the deal more attractive.

The involvement of major global banks like Goldman Sachs, JPMorgan, and Mizuho is believed to suggest confidence in SoftBank’s overall creditworthiness and the long-term value of its AI investments, despite the challenges of valuing the specific OpenAI stake.

As companies race to build infrastructure and develop new technologies, traditional funding sources are being stretched, leading to innovative financing structures and greater scrutiny from lenders.

However, a successful large-scale loan backed by a private AI stake might encourage similar deals, helping to channel more capital into the sector.

Conversely, if the financing falls through again, it could signal continued caution among banks toward private technology investments.

Palantir Technologies Chief Executive Alex Karp has delivered one of his strongest critiques yet of the commercial strategies adopted by leading artificial intelligence developers OpenAI and Anthropic, arguing that their token-based pricing models have become disconnected from what enterprise customers actually need as AI deployment costs continue to surge.

Speaking to CNBC’s Squawk Box on Wednesday, Karp said businesses are moving beyond measuring AI usage by token consumption and are instead demanding clear financial returns from their investments, a shift that could reshape competition across the rapidly evolving AI industry.

“I’m not throwing shade at them, but something has gone completely wrong,” Karp said. “The basic view among enterprises in this country is I’m going to chillax and waste my time with tokens.”

Karp’s remarks come as businesses grapple with the rapidly increasing cost of deploying advanced AI systems.

Successive generations of frontier models have become substantially more expensive to operate because they require greater computing power, larger graphics processing unit (GPU) clusters, and more sophisticated infrastructure.

As a result, many companies are moving away from optimizing token consumption, an approach often referred to as “tokenmaxxing,” and instead evaluating AI projects based on measurable business outcomes and return on investment.

The shift is prompting enterprises to reconsider whether relying on proprietary models from companies such as OpenAI and Anthropic offers sufficient value relative to their costs. Instead, organizations are now exploring open-weight AI models, which provide access to model parameters and can often be customized for enterprise use while operating at significantly lower costs.

Chinese AI Adds Competitive Pressure

The growing adoption of cheaper AI alternatives comes as Chinese developers continue narrowing the technological gap with leading U.S. AI companies. Open-source and open-weight Chinese models have become increasingly capable while remaining considerably less expensive to deploy, raising concerns within the U.S. technology sector that China’s AI ecosystem could erode the commercial advantage long enjoyed by American frontier laboratories.

Karp warned that the industry should not underestimate the pace of Chinese innovation.

“The industry should not underestimate the speed at which China is making progress in building AI models,” he told CNBC.

His comments echo broader concerns in Washington, where policymakers have intensified scrutiny of AI competition as Chinese firms rapidly improve their models despite U.S. restrictions on advanced semiconductor exports.

Many enterprises are now investing in proprietary AI systems tailored to their own operations rather than relying exclusively on third-party frontier models. Custom-built models allow companies to optimize performance for specific business tasks while maintaining tighter control over costs, intellectual property, and sensitive data.

That strategy aligns closely with Palantir’s own business model.

Earlier this week, the company expanded its partnership with Nvidia, combining Nvidia’s AI infrastructure and software with Palantir’s data integration platform to develop customized AI systems for U.S. government agencies.

The collaboration reflects growing demand for AI platforms that organizations can own, manage and adapt internally instead of depending entirely on externally hosted models.

Karp framed the debate as one of technological independence rather than simply pricing.

On Tuesday, Palantir published a nine-point “AI sovereignty” manifesto on social media platform X, noting that governments and enterprises should maintain ownership of their data, computing infrastructure and AI models.

The document criticized “tokenmaxxing” as a flawed commercial approach while encouraging organizations to retain control over the full AI technology stack.

According to Karp, enterprise customers want ownership rather than dependence.

“What aligns me with Nvidia, and I think is what the technical customers want, which is control over their compute, their models, their data stack and their alpha,” he said.

“They want to know they own the means of production. It’s not being transferred to someone else.”

The concept of AI sovereignty has gained traction as organizations seek to reduce reliance on external providers amid concerns over data privacy, cybersecurity, regulatory compliance and long-term operating costs.

Investors appeared to welcome Palantir’s positioning within the evolving AI industry. The company’s shares rose 8% on Wednesday, extending gains as investors increasingly see Palantir as a beneficiary of enterprise demand for customized AI deployments rather than generalized consumer-facing models.

While OpenAI and Anthropic continue to dominate frontier model development, enterprise customers are becoming more selective about how they deploy artificial intelligence. They’re now placing greater emphasis on measurable productivity gains, ownership of critical AI infrastructure, and long-term cost efficiency instead of simply accessing the most advanced models available.