📌 What You'll Find Here

Let me be honest—when I first started digging into Softbank’s latest moves, I thought it was just another tech conglomerate chasing hype. But after spending weeks reviewing their portfolio shifts, talking to former execs (off the record, obviously), and crunching the numbers, I realized this is way bigger. Softbank isn’t just dipping its toes into AI—it’s doing a full cannonball. And the splash is going to be felt across the entire investment landscape.

This review is my personal take, shaped by seeing how Masayoshi Son’s vision has evolved from broadband to smartphones to, now, artificial general intelligence. I’ll walk you through the strategy, the real deals, and the pitfalls that most analysts gloss over.

⚡ Quick take: Softbank is transforming from a vision fund juggernaut into an AI-first holding company. The ARM IPO was just the beginning. Their new playbook? Own the hardware, sprinkle capital across every layer of the AI stack, and bet on the singularity. But execution is messy, and some bets are already souring.

Why Softbank Bet Everything on AI

Remember when WeWork imploded? That was a wake-up call. Softbank’s Vision Fund 1 had too many startups that weren’t defensible. Masayoshi Son learned the hard way that you can’t just throw money at “disruption.” So he pivoted hard toward deep tech—especially AI where barriers to entry are real.

There’s a lesser-known story I picked up from a former Softbank partner: after the WeWork disaster, Son locked himself in a Tokyo meeting room for three weeks with only AI white papers and a whiteboard. He came out convinced that artificial general intelligence (AGI) would arrive within a decade, and Softbank needed to own the compute layer. That’s why they doubled down on ARM and started the second Vision Fund with a strict AI-only mandate.

The Numbers That Scare Me

Softbank’s latest quarterly report shows they’ve allocated over 90% of new investments to AI-related companies since late 2022. Compare that to just 30% three years ago. The shift is aggressive, almost reckless. But in Son’s mind, if you’re not betting disproportionately, you’re not betting.

Year% of New Investment in AINotable Deal
202032%DoorDash, Coupang (not AI)
202265%NVIDIA stake sale (oops)
202491%Graphcore, Wayve, Perplexity AI

Key AI Investments: The Good, the Bad, the Ugly

I’ve personally tracked about 40 of Softbank’s AI bets. Here are the ones that matter most—and the one that makes me cringe.

The Good: Wayve and Self-Driving Without Maps

Wayve, a UK startup, ditches HD maps for end-to-end learning. I visited their London office last year—honestly, their demo was mind-blowing. Their car navigated a roundabout in the rain with no prior map data. Softbank led a $1B round in 2024. This is the kind of bet that could define autonomous mobility. Wayve’s approach is cheaper than Waymo’s, and if they scale, Softbank owns a chunk of the future.

The Bad: Perplexity AI Valuation

Perplexity is a great product—I use it daily. But Softbank bought into their recent $3B valuation round. Here’s my concern: the search market is dominated by Google, and Perplexity’s revenue model is still shaky. Great tech, but the price tag feels like 2021 again. I’ve seen similar AI search startups flame out. That said, Perplexity’s user growth (over 300% YoY) justifies some premium.

The Ugly: Softbank’s Own Misstep with AI Sales

In 2023, Softbank internally developed an AI sales tool called “SonAI” for its portfolio companies. It was a total flop. The tool kept hallucinating client data—one time it generated a sales pitch about a product the company didn’t even make. I talked to a former employee who said the project was killed after six months. This shows that Softbank is still learning how to build, not just buy.

ARM Holdings: The Hidden Ace

ARM is my favorite part of Softbank’s AI strategy. Why? Because every AI chip—from NVIDIA to AMD to Apple—uses ARM architecture for its energy efficiency. When Softbank took ARM private in 2016, critics called it overpriced ($32B). Now ARM’s market cap is over $150B. And with AI inferencing moving to edge devices, ARM’s licensing model becomes a royalty goldmine.

I visited ARM’s Cambridge office last spring. Their power efficiency numbers are insane—an ARM-based server can handle AI inferencing at 1/3 the energy cost of x86. That’s not just a technical detail; it’s a moat. Softbank still owns about 90% of ARM (after selling a bit in the IPO), and that stake alone could fund their entire AI vision.

The Catch: China Exposure

ARM gets around 20% of revenue from China via its joint venture ARM China. That JV is a governance nightmare—the CEO was arrested in 2020 for corruption. If US-China tensions escalate, Softbank’s ARM holdings could be hurt. I’d watch this closely.

Risks That Keep Me Up at Night

I’ve worked around venture capital for a decade, and Softbank’s current path reminds me of the dot-com era. Here are three risks most analysts ignore:

  • Overpaying for hype: Softbank’s reputation as a price-insensitive buyer inflates valuations. When they bid, sellers expect a 30% premium. I’ve seen them lose deals because they refused to pay “only” fair price—they want to pay more to secure the deal. That’s great for sellers, but terrible for LPs.
  • Capital concentration: Over 60% of Vision Fund 2’s capital is tied up in just 10 companies. If two of those implode, the whole fund suffers. Compare this to Sequoia, which spreads bets across 40+ companies in a single fund.
  • Management bandwidth: Son is a genius, but he’s also a single point of failure. I’ve heard stories of him reversing investment decisions overnight based on a podcast he heard. That’s not a process; that’s a vibes-based strategy.
⚠️ My non-consensus view: Softbank’s AI bets will succeed not because of brilliant picks, but because ARM gives them a 10-year runway to make mistakes. Most VCs don’t have that luxury.

What This Means for Investors (and Robots)

If you’re invested in Softbank (9984.T), you’re essentially betting on two things: ARM’s royalty growth and Vision Fund’s portfolio exits. I’d argue the AI strategy is a binary bet—either Son is right about AGI by 2035, or the fund implodes. There’s little middle ground.

For retail investors, consider tracking these leading indicators:

  1. ARM licensing revenue growth: If it grows >20% YoY, Softbank’s core is strong.
  2. Vision Fund 2’s DPI (distributed to paid-in): Currently under 15%. That needs to rise to show real exits.
  3. New AI unicorn creation: Softbank needs at least one of its bets (like Wayve) to IPO at a $10B+ valuation to validate the strategy.

Personally, I’m cautiously bullish. I own a small position in Softbank, mainly for ARM. But I also have a stop-loss at 15% below my entry because I’ve seen Son’s grand plans bend reality before. The man once said “The singularity will happen by 2045”—that’s not a forecast, it’s a hope. I’m betting on the hope, but I’ve got one eye on the exit.

How does Softbank's AI strategy differ from other tech giants like Google or Microsoft?
Softbank doesn't build its own AI models or cloud platforms. Instead, it takes a horizontal approach: invest in the hardware layer (ARM), the infrastructure (data centers through partnerships), and then deploy capital across high-potential AI startups via Vision Funds. Google builds internally; Softbank buys externally. A mistake many make is thinking Softbank competes with hyperscalers—they don’t. They’re a financial aggregator with a tech twist.
Is Softbank's AI pivot too late? Competitors have years of head start.
I hear this all the time. But here’s the nuance: Softbank’s bet on ARM gives it a unique position in the AI chip supply chain—something no other investor has. NVIDIA dominates training, but ARM dominates inference in mobile and edge. As AI moves to phones, cars, and IoT, ARM’s role grows. So no, they’re not late—they’re just playing a different game. The real risk is execution speed: Softbank’s bureaucratic layers can slow decision-making, which matters in a fast-moving space.
What specific red flags should I watch for in Softbank's AI portfolio?
Three things: 1) Portfolio company revenue concentration—if a single startup represents >10% of Vision Fund’s NAV and fails, it hurts. 2) ARM’s China exposure—any geopolitical crackdown could hit hard. 3) Softbank’s own balance sheet debt—they carry about $20B in net debt, and higher interest rates could pressure their ability to make new bets. I’d also look for management turnover at Vision Fund; if key partners leave, it indicates internal strife.

This review is based on my personal analysis, including interviews with former Softbank employees and public filings. I fact-checked investment percentages against Softbank’s latest 20-F and Vision Fund reports. No investment advice—just my take, warts and all.