SanDisk-Kioxia's $31B Japan Bet Signals a New Memory Cycle

The Number That Changes Everything

$93.9 billion. That is the figure sitting on SanDisk’s balance sheet right now, representing contracted revenue that management treats as a floor rather than a hope. Two post-quarter deals pushed remaining performance obligations from $59.8 billion to that level, and the company is now betting the memory business will finally shed its legendary boom-bust volatility.

The $31 billion Japan expansion announced with Kioxia on August 27 is the physical expression of that bet. Part of it flows to Kioxia’s Kitakami plant, where an $11.3 billion facility will expand production of 10th-generation BiCS Flash — the 3D NAND technology both companies co-developed. The rest is a signal: after decades of treating memory as a commodity where volume decides everything, SanDisk is building a business that locks customers in for more than four years at a time.

For the stock, which has climbed more than 500 percent in 2026 alone, the question is whether this marks the beginning of a durable cycle or the climax of the last one.

The Cycle Breaker

The memory chip industry has always been a cruel teacher. Prices explode when demand outstrips supply, then collapse as every competitor builds the factory that floods the market. Samsung, SK Hynix, Micron, Kioxia — they all lived through it. The pattern is so reliable that analysts use the word supercycle almost reverently, knowing it usually means one more wave before the crash.

What makes this cycle different is the contract underneath it. Eight data center and edge customers signed New Business Model agreements with pricing floors and ceilings. That structure changes how the business behaves. When demand surges, SanDisk cannot simply raise prices to capture windfall margins. When demand falls, the floor prevents a free-fall. The companies have essentially traded upside optionality for downside insurance, and management says that trade will deliver mid-teens growth in both revenue and profit with margins holding near 80 percent.

The numbers support the thesis, at least for now. Data center revenue surged 437 percent year over year to $5.2 billion in fiscal 2026, with the fourth quarter alone nearly doubling sequentially to $2.9 billion. Edge products — PCs and phones running AI features — grew 195 percent to $12.2 billion. Total revenue hit $20.2 billion, up 175 percent, and the company swung from a $1.6 billion net loss to $11.4 billion in net income in a single fiscal year.

AI is draining memory supply faster than the industry can replace it, and that imbalance is real. But the more interesting story is the institutional change: long-term contracts replacing spot-market pricing as the default way to sell memory chips.

Why Japan Matters Beyond Geopolitics

China watchers will see the Japan investment and immediately think about export controls and self-sufficiency. And they are not wrong. Every major memory chip expansion outside Chinese territory reinforces the strategic decoupling that Washington and Tokyo have been engineering. But the geographic argument obscures something more important about what is happening at Kitakami.

Japan has spent decades losing ground in semiconductors. Kioxia — formerly Toshiba Memory — was spun off in 2017 as a symbol of how far the country had fallen. The $31 billion bet, discussed with Prime Minister Sanae Takaichi on the same day as the announcement, is an attempt to reverse that narrative. It is also a remarkably expensive one.

For an international audience that tends to view semiconductor investment through the lens of Taiwan and South Korea, the Japan move deserves attention because it shows where capital is flowing when the AI boom demands memory faster than existing supply chains can deliver. SanDisk already had relationships with Kioxia through their joint BiCS development. This expansion formalizes that partnership with unprecedented scale. The companies are not testing the water. They are building a larger pool.

The Buyback Warning

Here is where the story gets complicated. SanDisk management has committed to returning nearly all excess cash to investors through buybacks, adding $14 billion to the authorization to bring the total to $15.5 billion — roughly 7 percent of the company’s market value. The language from earnings calls sounds disciplined. The execution does not.

Buybacks only create value when executives buy below intrinsic value. There is no indication SanDisk’s leadership is doing that calculation. The commitment to send back nearly all excess cash regardless of price is closer to indiscriminate purchasing than capital allocation. When a company that has rallied 500 percent in a year announces a massive buyback program, the market should ask whether the shares were a bargain six months ago — and whether they are a bargain today.

This is not a new criticism of the memory business. Micron has faced similar questions during previous upcycles. But the scale matters. SanDisk is running an $11.4 billion profit year and planning to return a substantial portion of it to shareholders while simultaneously spending $31 billion on new capacity. Those are not contradictory decisions, but they require a precise assessment of where the cycle sits that the current buyback stance does not reflect.

What Comes Next

The contracted revenue floor of $93.9 billion gives SanDisk a runway that previous memory cycles did not enjoy. If the contracts hold and data center demand continues its current trajectory, the company has a multi-year basis for steady growth rather than volatile swings. The risk is not that AI demand disappears. The risk is that the industry, seeing these numbers, decides to build anyway — and then the contracts expire or the pricing floors prove insufficient to prevent a price war.

The Japan expansion timeline extends through 2032. That gives the market three years to test the thesis before the full cost of the investment shows up on the income statement. If memory prices remain elevated and the contracted revenue base holds, SanDisk will have built the most durable memory business in the sector’s history. If the cycle peaks before 2032, the buyback program and the capital expenditure will look like the excesses of a company that mistook a supercycle for a permanent regime change.

The stock is up 500 percent. The contracts are real. The bets are large. The memory industry has survived every cycle before this one. The question is whether the contracts make this one different — or whether the market has simply forgotten how cruel the teacher can be.