One deck, two numbers: an all-time high and ¥14 trillion gone

Here's the deal: on August 6, SoftBank Group reported Q1 FY2026 results for April through June. The headline is spectacular. Net asset value of ¥72.3 trillion as of June 30 — roughly $445.2 billion — the highest in the company's history. The largest single contributor was Arm Holdings at ¥43.4 trillion, boosted by an Arm share rally of more than 17% shortly before the print.

The same materials carried another number. Pro forma NAV as of August 5: ¥58.3 trillion. Roughly ¥14 trillion evaporated in five weeks. A tech selloff intervened, and nearly all of SoftBank's assets are marked to market. SoftBank Group shares themselves fell over 4% on the day.

Those two numbers sitting on the same slide are an accurate portrait of SoftBank in 2026. This is no longer an operating company so much as a leveraged position on AI assets. And Masayoshi Son keeps making it bigger.

The asset list matters more than the income statement

Start with earnings anyway. Q1 net income came in at ¥347.3 billion, down 18% year over year, though ahead of consensus. Investment gains were ¥1.8 trillion, up ¥1.3 trillion from a year earlier, driven mainly by mark-ups on Intel and ByteDance. Net income fell on rising costs and flat valuation on some holdings — several outlets specifically flagged that OpenAI's carrying valuation didn't move much this quarter.

Worth stating plainly: SoftBank Group's quarterly net income is nearly meaningless as an operating metric. Most of the profit comes from mark-to-market swings on holdings, which move by trillions of yen quarter to quarter. That's precisely why Son has spent years telling people to look at NAV instead. So the real content of this print sits in the asset list and the balance sheet metrics, not the P&L.

Item Detail
Reported August 6, 2026 (Q1 FY2026, Apr–Jun)
Net asset value ¥72.3T (~$445.2B) — all-time high, as of June 30
Pro forma NAV ¥58.3T as of August 5 — down ~¥14T in five weeks
Largest holding Arm Holdings, ¥43.4T
Net income ¥347.3B, −18% YoY (beat consensus)
Investment gains ¥1.8T (up ¥1.3T YoY), led by Intel and ByteDance
OpenAI additional $20B (~¥3.2T) deployed in April and July
OpenAI cumulative $44.6B
OpenAI planned $10B more in October → cumulative $64.6B
Loan-to-value 13% (policy ceiling 25%)
Cash ¥2.3T — covers 2+ years of bond redemptions
Share reaction About −4.4%

The line that stands out most is OpenAI. Twenty billion dollars deployed across April and July brought the cumulative position to $44.6 billion, and another $10 billion planned for October takes it to $64.6 billion. That is overwhelming by the standards of SoftBank's own history. For comparison, the Arm acquisition in 2016 was roughly $32 billion. SoftBank is putting twice the price of Arm into a single startup.

The second standout is LTV at 13%. With a policy ceiling of 25%, the company is running at about half its self-imposed limit, and ¥2.3 trillion of cash covers more than two years of bond redemptions. The reason SoftBank emphasizes these figures is obvious: as the bets get bigger, so does the question "are you gambling with borrowed money?" This is the preemptive answer. But note that LTV has asset value in the denominator — when NAV goes from ¥72T to ¥58T, the same debt produces a higher ratio. Read the ¥14 trillion and the 13% together.

Who gets what out of this structure

SoftBank Group is going for exposure across multiple layers of the AI value chain simultaneously: silicon design (Arm), models (OpenAI), and the applications running into data centers and robotics. The phrase Son has repeated for years is an "artificial superintelligence" ecosystem, which in practice is less a bet on any company than a strategy of buying every asset class that must appreciate if AI is real. That this quarter's gains came from Intel and ByteDance shows how wide the net is.

OpenAI gets capital and time. Frontier model training and inference infrastructure demand tens of billions of dollars per year, and fewer than ten pools of capital worldwide can write those checks. At $64.6 billion, SoftBank would be the largest single investor apart from Microsoft. For OpenAI, capital that isn't attached to a specific cloud provider carries real strategic value.

Arm functions as the collateral in this structure. Of ¥72.3 trillion in NAV, ¥43.4 trillion is Arm — meaning roughly 60% of SoftBank's assets sit in one listed semiconductor IP company. That concentration cuts both ways. Arm up means NAV up means more capacity to deploy; Arm down means the reverse. The ¥58.3 trillion figure as of August 5 is that mechanism working in real time.

There's a Japanese market dimension too. SoftBank Group is a top-weighted name on the Tokyo exchange, and its NAV swings are transmitted into Japanese index performance. It's one of the channels through which a US tech selloff reaches Japanese equities — structurally similar to what two semiconductor names do to the KOSPI.

Intel is the quiet protagonist of this quarter. Intel and ByteDance being named as the main contributors to ¥1.8 trillion of investment gains means SoftBank booked substantial mark-ups on its Intel position. What's notable there is the change in portfolio character. Son's archetypal investment is a high-growth private startup; Intel is the opposite — a large listed company in the middle of a restructuring, which is closer to a value position. That a large share of this quarter's gains came from that kind of holding says the portfolio is more varied than its reputation, and also illustrates how much quarterly results depend on which marks happened to move.

Conversely, SoftBank shareholders still carry an old problem. The stock has historically traded at a substantial discount to NAV, for reasons including holding-company discount, illiquidity of the assets, and above all the governance risk of concentrating capital allocation in one person's judgment. A record NAV doesn't fully translate into share price, and that structure hasn't changed.

The chronic condition: the NAV discount

There's one more thing you have to know to understand SoftBank. Its market capitalization has traded well below NAV for years. The company can say "our assets are worth ¥72.3 trillion" and the market will still assign a far lower number. Son emphasizing NAV every quarter is an attempt to close that gap, and repeated buybacks serve the same purpose.

The discount has three sources. First, the ordinary holding company discount — a shareholder could simply buy Arm shares directly instead of routing through SoftBank, and the intermediate layer adds headquarters costs and taxes. Second, liquidity. If SoftBank sold its Arm stake into the market, the price would collapse, so ¥43.4 trillion on the books is not ¥43.4 trillion in cash. For a private asset like OpenAI, the problem is much larger. Third, capital allocation risk: shareholders cannot control whether cash from asset sales is returned to them or funneled into the next bet. And on this company's record, the answer has usually been the latter.

This quarter's print reinforced that third item rather than easing it. Twenty billion dollars already deployed and ten billion more scheduled for October is an unambiguous signal. When assets rise, SoftBank doesn't sell — it buys more. The structure where one person's conviction is the sole criterion for capital allocation is fully intact, which is why a record NAV doesn't lift the share price proportionally.

There are scenarios where the discount narrows: OpenAI going public so the private asset gets a market mark and a liquidity path, or the company executing a large buyback or returning proceeds after asset sales. Which is why, for SoftBank shares, the news that matters more than earnings is buyback announcements and progress toward an OpenAI listing.

Alibaba returned thousands of times. WeWork went to zero.

SoftBank's investment record contains both extremes. The signature success is Alibaba: roughly $20 million invested in 2000, a stake later worth tens of billions. That single position justified the company's existence for two decades. The lesson is structural — Son's approach is venture arithmetic where dozens of positions can go to zero if one returns a thousand times. And it worked once, for real.

The signature failure is WeWork. SoftBank put in more than $10 billion, and a $47 billion valuation ended in a failed IPO and bankruptcy. The lesson there is different. WeWork didn't fail because the idea was bad — it failed because the investor manufactured the valuation. SoftBank kept investing at successively higher prices, creating an internal mark rather than a market price, and everything collapsed the moment the outside market refused that price. That's exactly why people invoke WeWork when they see repeated OpenAI investments. But there's a decisive difference: WeWork's losses grew faster than its revenue; OpenAI's revenue is growing steeply.

Third case: the 2000 dot-com collapse, in which SoftBank lost more than 90% of its market value and nearly died. That experience is the root of today's financial policy — the 25% LTV ceiling, the two-year redemption cash buffer. So when the company emphasizes 13% LTV and ¥2.3 trillion of cash, read it not as marketing but as the rules written by a company that has already died once.

Fourth: Vision Fund. Launched in 2017 at roughly $100 billion, once the largest venture fund on Earth, it posted record losses in fiscal 2022. The lesson is about the relationship between capital and timing. Too much capital pushes you toward large deals rather than good deals, which inflates your own entry price. Whether the $64.6 billion OpenAI position avoids that trap can be verified only one way: whether OpenAI's next external round prices above the current mark.

How other pools of capital move

Microsoft remains OpenAI's most important partner, and its money comes bundled with cloud — a fundamentally different instrument from SoftBank's. SoftBank writes pure capital; Microsoft writes capital plus infrastructure commitment. The distinction matters to OpenAI: the more pure capital it has, the more leverage it holds over infrastructure suppliers.

Middle Eastern sovereign funds are the quiet whales here. Vehicles including the UAE's MGX are deploying at scale into AI infrastructure and frontier labs, and unlike SoftBank they face no quarterly public-market scrutiny. Their cost of capital and time horizon are more favorable. Part of why Son keeps scaling is this competition.

Nvidia occupies an unusual position as both capital provider and supplier. Taking equity stakes in companies that buy its chips invites circular-financing criticism, but functionally it accelerates the whole ecosystem. For SoftBank to play in that layer it needs semiconductor assets, and Arm is exactly that.

Traditional venture capital is getting priced out of this competition. When a single frontier lab round runs to tens of billions, a multi-billion-dollar fund becomes a name on the cap table. That's why sovereign funds, crossover funds, and corporate capital — not VCs — have become the principals in recent AI financing, with SoftBank as the original archetype.

The last variable is whether OpenAI goes public. Realizing value on a $64.6 billion position eventually requires a liquidity path: an IPO or a large secondary. Part of why the market discounts SoftBank shares against NAV is precisely this liquidity uncertainty, so progress on an OpenAI listing could narrow that discount. Delay keeps it wide.

So what actually changes

For general readers, nothing directly. But this print makes the scale of the AI capital cycle concrete. One company putting $64.6 billion into one startup means capital at the level of a national budget is flowing into private AI — and a large share of it ends up in GPUs and data centers.

For startups and investors, it's a signal about the financing environment. Late-stage AI fundraising has moved beyond traditional VC into SoftBank, sovereign funds, and corporate capital. That money is generous on valuation but complex on terms and different on time horizon. For an early-stage founder, the practical meaning is modest: abundant late-stage capital means there's somewhere for your company to sell into later.

For investors, it's a reminder of what SoftBank the security actually is. It moves on mark-to-market values of holdings rather than operating results, 60% of those assets are one company, and the largest new position is a private company. In other words, a leveraged AI index with an illiquidity premium attached. NAV losing ¥14 trillion in five weeks is that structure's volatility on display. At the same time, 13% LTV and two years of redemption cash are the shock absorbers built for it.

For Korean readers, there's a useful comparison. SoftBank's role in Japanese indices structurally resembles what SK Hynix and Samsung do to the KOSPI — an index tied to a few names, and those names tied to overseas AI demand. That both markets wobbled for related reasons in the first week of August isn't a coincidence.

One sentence: SoftBank recorded the largest asset value and the largest single bet in its history in the same quarter, and the fact that those two move together is both its only risk and its only opportunity.

🥄 Three Things You're Probably Wondering

— So what does this mean for me? Little directly. But a large share of the $64.6 billion SoftBank is putting into OpenAI ends up in GPUs and data centers. If you hold AI infrastructure names or funds, it's useful to know the size of that flow.

— Why now? An Arm rally of more than 17% pushed June 30 NAV to an all-time high right as the quarterly print landed. Meanwhile the late-July/early-August tech selloff had already taken pro forma NAV down ¥14 trillion. Having both numbers in the same deck is what made this report a talking point.

— Is $64.6 billion reckless? Too early to call. The decisive difference from WeWork is that OpenAI's revenue is genuinely growing fast. The WeWork pattern — one investor repeatedly marking up its own position and manufacturing an internal valuation — remains a live risk, and the test is whether the next external round prices above the current mark.

Sources

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!