Latest / International Business with Fexingo: Global Trade, Cross-Border Deals, and Multinational Operations / How SoftBank Is Becoming the Global AI Infrastructure Banker
Transcript
- Lucas: SoftBank shares surged more than 11 percent yesterday, and the rally is carrying into this morning. That single move added something like 15 billion dollars to its market cap. The question is whether this is just another momentum pop or something more structural. Luna: Eleven percent is eye-catching. But SoftBank has had these moments before — the Alibaba IPO, the ARM sale, the WeWork disaster. What's different this time? Lucas: What's different is the story underneath. SoftBank is reinventing itself as what I'd call the world's first dedicated AI infrastructure banker. Not a venture fund placing scattered bets — a financial institution that originates, structures, and holds loans backed by physical AI assets. Data centers, GPUs, power connections. Luna: That's a big pivot from the Vision Fund model. And honestly, if a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, that's literally what funds making episodes like this possible. No ads, no sponsors — just people who find the show useful. So thank you if you're one of them. Lucas: Yeah, genuinely. Keeps us independent. Anyway — back to SoftBank. The pivot makes sense when you look at where the money is flowing. The trade-weighted dollar index has softened a bit, down to 120.7, and the yen has strengthened slightly to 160.9 per dollar. That's a tailwind for Japanese exporters and for a company like SoftBank that holds a lot of dollar-denominated assets. Luna: And the broader Asian tech rally is helping. The FXI, the China large-cap ETF, is up almost 5 percent in the last five days. SoftBank rides that wave too. Lucas: Exactly. But the core thesis is about the asset class itself. SoftBank has committed roughly 100 billion dollars to AI infrastructure over the next few years. That includes the massive data center projects in the US, UK, and Japan. They're also setting up a dedicated lending arm that makes loans secured by GPU clusters. Luna: gpu backed loans. That is fascinating. So instead of buying equity in AI startups, they're providing the debt financing for the hardware those startups need to operate. Lucas: Right. The asset has intrinsic value — NVIDIA H100s, B200s, the next generation. They don't depreciate as fast as people think if they're under contract. So SoftBank can lend at attractive rates, collect interest, and if a borrower defaults, they take possession of the GPUs and re-lease them. It's a secured lending business with real collateral. Luna: That sounds much more predictable than the old Vision Fund model, where you'd write a check to a company like WeWork and hope the valuation went up. Lucas: Much more. And the irony is that SoftBank is becoming the bank it always wanted to disrupt. Masayoshi Son has talked for years about the Singularity, about AI changing everything. But the business model that's actually working is plain-vanilla asset-backed lending — just at enormous scale and with a very specific asset class. Luna: What does this mean for cross-border capital flows? Because 100 billion dollars doesn't stay in one country. Luna: That's a good point. A lot of that capital is flowing into the US — Texas, Virginia, Ohio — but also into the UK and Southeast Asia. SoftBank is essentially recycling Japanese capital — cheap yen — into dollar-denominated infrastructure assets. That's a classic carry trade, but with a real economic output attached. Lucas: And it changes the risk profile for the countries receiving that capital. Data centers are long-duration investments. They create jobs, they strain local power grids, they require regulatory approvals. SoftBank isn't just a passive investor — they're becoming a developer and operator. Luna: There's a headline this morning about SK Hynix's potential Wall Street IPO and whether it can narrow the 'South Korea discount.' SoftBank is a major customer for SK Hynix's high-bandwidth memory chips. Lucas: They are. And if SoftBank is financing data centers that buy those chips, you have a vertical integration story without the vertical integration. SoftBank doesn't own SK Hynix, but its lending business creates demand for their products. That's a symbiotic relationship that benefits both. Luna: Let's talk about the risks. Because 11 percent rallies don't happen without a downside. Lucas: The biggest risk is that the AI buildout slows. If corporate spending on AI models decelerates, the demand for GPU compute drops, and those loans start looking less secure. There's also the risk that interest rates stay higher for longer — that raises SoftBank's cost of capital and squeezes margins on their lending. Luna: And regulatory risk. Data centers consume enormous amounts of electricity. In some parts of the US and Europe, utilities are struggling to keep up. If governments start capping new data center connections, SoftBank's pipeline gets constrained. Lucas: That's already happening in places like the Netherlands and Singapore. SoftBank has had to pivot some projects to Malaysia and Indonesia. But that also opens up new geographies. The point is, the thesis is not without execution risk. Luna: What about the old portfolio? SoftBank still holds Alibaba shares, ARM, a bunch of Vision Fund stakes. Are they unwinding those to fund this new strategy? Lucas: They've been selling Alibaba gradually for years. ARM is the crown jewel — they still own about 90 percent. But the new lending business is separately capitalized. SoftBank raised debt — about 10 billion in bonds earlier this year — specifically earmarked for AI infrastructure. They're not fire-selling assets to fund it. Luna: So it's additive, not a replacement. Lucas: Additive and strategically coherent. Son's vision has always been to be the platform for the AI revolution. The Vision Fund was the equity side — take stakes in all the companies building AI. This lending arm is the debt side — finance the physical infrastructure those companies need. Together, they create a kind of full-stack AI finance machine. Luna: And the market is rewarding that clarity. The 11 percent rally suggests investors see a real business here, not just a narrative. Lucas: Exactly. The question now is scale. Can SoftBank deploy that 100 billion efficiently? Can they maintain underwriting discipline when everyone wants a piece of AI? If they can, they'll have built something genuinely new — a bank for the age of intelligence. Luna: And if they can't, we'll be having a very different conversation in two years. Lucas: We will. But for now, the market is giving them the benefit of the doubt. And that's rare for a company that has burned so much goodwill over the past decade.