Summary
- SoftBank Group pledged up to €75 billion ($87 billion) to build AI data centers in France - the biggest infrastructure bet of its kind in Europe. The market rallied the stock. I believe the headline number obscures a much harder question: where does the money actually come from?
- SoftBank's free cash flow was negative $15.2 billion in fiscal year 2025. Net debt sits around $95 billion. This pledge is funded by borrowing, not cash generation.
- The Vision Fund booked a $46 billion gain last year driven almost entirely by SoftBank's unrealized stake in OpenAI. Paper gains are not deployable capital - you can't pay construction crews with valuations.
- Europe's power grid is already strained. Data center projects face decade-long interconnection queues and rising electricity prices that are pushing work toward cheaper markets. SoftBank's 3.1 gigawatt first-phase target collides with that reality.
- The rally rewards a narrative, not structural fundamentals. I rate SoftBank as a Hold. The strategic direction is correct, but the execution risk is extraordinary.
I've been very surprised that SoftBank Group shares rose following the announcement of its massive French data-center plan. The market treated a €75 billion pledge as proof of strategic clarity and conviction. In my opinion, it's evidence of something else: a company betting its entire financial architecture on the assumption that AI demand will never slow and that borrowing capacity is infinite.
A commitment letter is not a balance sheet.
The numbers tell a different story from the headline. SoftBank reported negative $15.2 billion in free cash flow for fiscal year 2025 - its worst year of cash consumption on record. Free cash flow is the cash left after a company funds all its operating expenses and capital expenditures. When it's negative, the company is spending more cash than its operations generate. You cannot fund a $52 billion infrastructure buildout from operations when your operations are consuming $15 billion a year.
That being the case, the pledge has to be financed. SoftBank's net debt - total debt minus cash on hand - already stands at roughly $95 billion. Adding even the first-phase €45 billion to that leverage profile would push total debt close to or beyond $140 billion. The company's balance sheet is already stretched. This plan requires either significantly more borrowing or asset sales to free up cash. Neither is trivial.
Here's where the narrative gets thinner. Last quarter, SoftBank's Vision Fund posted a $46 billion gain, driven almost entirely by rising valuations in its OpenAI stake. SoftBank also holds $56.7 billion in unrealised fair value from 384 private investments. These numbers look impressive on a press release. They are not cash. They are paper - valuations that can swing in either direction depending on market sentiment and the next funding round.
I'm not dismissing the OpenAI position. It's real, and if OpenAI achieves the market dominance its trajectory suggests, SoftBank's stake could be worth substantially more. But here's the structural problem: your best-performing asset is also your least liquid one. If SoftBank needs to draw down capital for the French buildout and the private market tightens, those paper gains become irrelevant to the funding gap.
That brings me to the second structural problem - one the market has barely priced in: Europe's power grid.
A 3.1 gigawatt data center deployment in France sounds manageable until you consider that European data center power demand is projected to double to 36 gigawatts by 2030, already straining grids across Germany, the UK, and France. New projects face decade-long interconnection queues. Facilities are being built at half capacity because they can't get enough power. A study published in May 2026 exposed the energy crisis at the heart of Europe's AI push: the continent simply cannot feed the machines fast enough.
SoftBank chose the Hauts-de-France region - specifically Dunkirk, Bosquel, and Bouchain - partly because the Port of Dunkirk has offered priority power connections. That's smart positioning. France's nuclear-heavy grid also gives it a cleaner energy profile, which matters for hyperscaler ESG commitments and regulatory favor. But a priority connection for 700 megawatts at Dunkirk does not solve the grid equation for 3.1 gigawatts spread across multiple sites. The local infrastructure has to scale too, and that takes years, not quarters.
CNBC reported in mid-May that data center projects are already migrating to parts of Europe with lower power costs, creating winners and losers across the continent. Electricity prices are redefining the geography of AI infrastructure. SoftBank is betting that France can keep up. That's a geopolitical and regulatory bet as much as a construction one.
The counterargument is straightforward: SoftBank has made mega-bets before, and the AI infrastructure market is growing fast enough that even a leveraged, slow-to-execute position can create value. The Vision Fund's pivot toward AI - including layoffs of 20% of employees to refocus on bold AI bets - signals genuine strategic realignment, not just headline theater. If 5 gigawatts of AI compute capacity eventually comes online in France, SoftBank would control one of the largest colocation and wholesale infrastructure footprints on the continent. The long-term revenue potential is real.
I don't dispute that. But conviction in the destination doesn't guarantee the bridge.
The false narrative here is that a large announcement number equals investment-grade execution. SoftBank's track record of mega-bets tells a more cautious story. The Vision Fund 2, launched in 2019 with $60 billion in committed capital, lost tens of billions over its early years before edging into the black. The pattern is familiar: overcommit, underperform, pivot, and hope the market rewards the new direction before the old losses crystallize.
This time the thesis is structurally sounder - AI demand is secular, not cyclical, and the cross-pollination across hyperscaler capex means the entire infrastructure stack benefits regardless of which chip architecture wins. I've written repeatedly that the AI infrastructure buildout is the most durable secular demand shift of this decade. SoftBank is positioning on the right side of that shift.
However, the gap between the pledge and the cash is the investment problem. Negative free cash flow, $95 billion in net debt, paper gains concentrated in a single private position, and a power-constrained build environment - these are not abstract risks. They are the conditions that turn ambitious plans into balance sheet overhangs.
That being the case, I rate SoftBank Group as a Hold. The strategic direction toward AI infrastructure is correct, and the France play leverages real advantages in nuclear power and regulatory alignment. But the capital structure required to execute is leveraged to a degree that the market has not yet priced in. Investors should wait for the first-phase financing to close, the initial construction milestones to materialize, and free cash flow to turn positive before treating this as anything more than an intention.
The market rewards conviction. My job is to check whether conviction is backed by cash. In this case, the answer is not yet.