GSI TECHNOLOGY INC (GSIT) stock intrinsic value fair value estimate & owner earnings analysis

GSI Technology makes specialized, very fast SRAM memory chips used in networking equipment, semiconductor test and emulation systems, and radiation-tolerant versions built for military, defense and aerospace programs. Customers include BAE Systems, General Dynamics, Rockwell and IBM on the defense/aerospace side, and Nokia, Cadence Design Systems and the Taiwan-based KYEC on the commercial side. Chips are designed in the U.S. and manufactured through a Taiwan-based wafer foundry; about 17% of fiscal 2026 revenue shipped to China, with the balance spread across the U.S., Singapore, Germany, the Netherlands and other markets. Alongside this legacy memory business, the company has spent several years and tens of millions of dollars developing Gemini, an associative-processing chip aimed at very fast AI search and pattern-matching. So far this program has produced only small, government-funded research awards - a little over $1.6 million from an Air Force Research Laboratory contract in fiscal 2026 - rather than commercial product sales. Owner earnings, meaning cash the business actually generates after real costs such as stock-based compensation and the capital spending needed just to maintain current operations, have been negative every year since fiscal 2021, averaging roughly $0.55 to $0.80 of loss per share annually for six straight years. That six-year stretch is when the current pattern took hold - a slowly eroding legacy memory line funding a large, so-far unproductive R&D program - and it is what we used to set the starting point for value, rather than fiscal 2019-2020, when the same math showed the company close to breakeven, before the heavy AI-chip investment began. One accounting point worth flagging: fiscal 2026's reported net loss looked better than fiscal 2025's, but that improvement came almost entirely from a $6.2 million non-cash paper gain on warrants issued in an October 2025 stock offering that raised roughly $50 million before expenses. Actual cash burned in operations was larger in fiscal 2026 than in fiscal 2025, and the first quarter of the new fiscal year showed the burn rate worsening further, not improving. We built our near-term outlook from that cash trend rather than from the flattering reported number, and the share count has grown by roughly 45% over the past year and a half as a result of this and other stock issuances. The legacy memory business also shows a specific, disclosed crack rather than just a general soft patch: Nokia, historically one of the company's largest customers, has been replacing these chips with its own internally developed memory in newer product designs, and Nokia's share of revenue has fallen from about 21% to about 6% over two years. The defense and aerospace portion of the business, which requires long and expensive qualification cycles, looks stickier, but the commercial telecom portion looks structurally, not just cyclically, at risk. Against that backdrop, all three cases we modeled show continued cash losses - the question is how much and for how long, not whether: Bear case: the Nokia-style substitution spreads to other customers, R&D spending is not reined in, and the company returns to the capital markets more than once over the next decade to keep funding the losses, roughly doubling the share count again. Base case: the defense and aerospace portion of the memory business holds up while the commercial telecom portion keeps shrinking, government research funding for Gemini continues at a modest pace, and losses gradually narrow over the decade as the company reapplies the cost discipline it already showed once - it cut headcount by about 16% in August 2024 when cash burn became a concern - though the AI-chip program never becomes self-sustaining within our ten-year window. Bull case: that same cost discipline is applied earlier and more fully, losses shrink to a stable $8-10 million a year within a couple of years, and the whole ten-year path is funded out of the roughly $80-90 million of cash already on hand without further share issuance. It is still a loss-making business in this case, just a smaller one. None of the three cases assumes Gemini becomes a meaningful commercial product, because we found no signed commercial contracts behind it, only small government research awards - so we did not build any case around eventual AI-chip success. Discounting these owner-earnings paths at a 6% rate over ten years, with nothing assumed to follow afterward because we found no evidence of a durable, profitable engine beyond that point, the ongoing operations come out worth less than zero per share in every case: roughly -$2.48 in the bear case, -$1.98 in the base case and -$1.89 in the bull case. That does not mean the company has no value at all - it holds close to $90 million in cash and no debt - but it does mean that, based on the business as it is actually being run today, continuing to fund the current spending plan destroys value rather than builds it, and on that basis the estimate supports no positive margin-of-safety price. A positive estimate would require the company to sign a real commercial contract for the AI chip, or to show several consecutive quarters of cash burn declining substantially toward breakeven.