GSI Technology: A Good SRAM Business Wearing a Speculative AI Multiple

I recommend a bearish rating on GSI Technology, Inc. (GSIT) with a fair value estimate of $3.25 to $3.80 per share versus the stock’s close of $5.63 per share on July 24, 2026. The company’s core SRAM business is undeniably improving, and the balance sheet is the strongest in its peer group, but the stock price is inflated based on the promise of an AI chip franchise that has yet to materialize, while the performance claims behind that franchise are currently the subject of a securities-fraud investigation.

A Falling Stock That I Think Is Still Too Expensive

GSIT has fallen nearly 69% from its peak of $18.15 per share on October 20, 2025. That kind of drawdown usually suggests a stock is priced for disaster, and my initial instinct was to look for a value opportunity. However, my due diligence uncovered a company that has declined for understandable reasons, has further downside based on my analysis, and is currently being buoyed by a story that has yet to produce a single dollar of revenue.

The more important context, in my opinion, is that GSIT jumped to $10.55 on May 26, 2026 after a clean fiscal Q4 report, then round-tripped nearly half of that gain away, closing at $5.63 per share on July 24, representing a 47% drop in just over two months with no negative news on earnings. Something else is weighing on this stock, and I’ll tell you what shortly.

GSI Technology designs and sells SRAM (static random-access memory) chips for networking, defense, aerospace, and test-equipment customers out of Sunnyvale, California, and is also developing the Gemini Associative Processing Unit, a compute-in-memory chip architecture the company is pitching for edge AI inference in drones, smart cities, and defense applications, per its fiscal 2026 Form 10-K. Two businesses, two very different risk profiles, and I think the market is currently pricing them as one.

The SRAM Business Is Actually Working

Fiscal 2026 (ended March 31, 2026) revenue grew 22.4% year over year to $25.1 million, driven by an increase in gross margin to 54.5% from 49.4% and a rise in defense sales to ~46% of total quarterly shipments. The growth is fueled by demand for rad-hardened, defense-grade SRAM for satellites and missiles, not AI inference chips, which is a niche market with long qualification cycles that creates a moat against competitors. This is the one area in which I would feel comfortable investing, given the technical credibility of the product.

The company posted a net loss of $13.25 million, or $0.42 per share for the year, but I would not put much weight on this figure due to the mark-to-market accounting for warrants and financing costs associated with the October 2025 raise, as detailed in the 10-K. A better metric is operating income, which saw a wider loss of $17.5 million from $10.8 million for the year. This compares unfavorably to the same period last year, which benefited from a one-time asset-sale gain, not likely to be repeated in 2026. Overall, once the one-off costs are taken out, the actual rise in spending is much lower, and most of it is tied to Plato R&D.

For fiscal Q1 2027, management expects revenue of $5.9 million to $6.7 million and a gross margin of 54-56%, indicating minimal change from the previous quarter. The results are expected to be released on July 30, 2026.

The Elephant in the Room: Is Gemini-II Real, or Is This a Story Stock?

Gemini-II has had several technical successes, including a proof-of-concept win with partner G2 Tech for a drone-surveillance program, a UUUnity Software Inc.$43.00+5.37% Valuation Verdict Overvalued For a complete valuation analysis, visit the Unity Software Inc. Stock Analysis page.Run Full Analysis.S. Army SBIR contract that has moved to Phase II, and a Taiwan smart-city pilot. However, each of these is a proof-of-concept, an SBIR phase, or a pilot, and none are revenue-generating. During the fiscal Q4 earnings call the company’s VP of Sales Didier Lasserre stated that the company was “still in the early stages” of commercializing Gemini and that the Plato chip, which is intended to be the production chip, would not be taping out until early 2027 with silicon availability in summer 2027, and revenue is unlikely before fiscal 2028.

In addition, according to the case page for Rosen Law Firm, on February 3, 2026, a user on Stocktwits published a post claiming that GSI was “almost certainly hiding” that its chip “did not run Gemma-3 at all, only the pre-generation RAG phase,” and that the APU “lack[s] the MAC units required for matrix multiplication.” Following this report, the stock closed at $6.52 on February 4, 2026, down $1.08. Rosen Law Firm and several other law firms are investigating whether GSI Technology disseminated misleading information to investors. As of the publication of this article, the investigation is ongoing. While this does not prove that the company committed any wrongdoing, the allegation is damaging to the value of the shares because it challenges the technical assertions made by the company’s management about its AI capabilities. As a result, I place a heavier discount on management’s forward technical claims when valuing the AI opportunity below.

Where I Land: SRAM Multiple Plus Net Cash Plus a Small, Discounted AI Option

One reason I want to skip the blended DCF is I don’t think I would be doing the math right since you’d be comparing two completely different types of businesses. Instead, I think it’s best to value this house like any conservative investor would – assume a peer multiple for the operating business, add cash dollar for dollar, and attach a probability weighted option value to the catalyst.

For the SRAM business, I figured a multiple ranging from 2.5x to 3x based on the $25.1 million of sales, with Netlist, Inc. (NLST) at the high end (3.08x) and Diodes Incorporated (DIOD) at the low end (2.59x). For this reason, I decided not to include GSIT’s own EV/sales multiple of 6.92x, since I don’t think that is based on an actual opportunity for an artificial intelligence business and so it is unfair as a data point for the calculation. This leaves the core enterprise value at $62.8 million to $75.4 million.

If I add in net cash of $58.75 million, which is $67.2 million of cash minus $8.47 million of total debt, consisting almost entirely of operating-lease liabilities rather than bank debt, I have a base equity value of $121.5 million to $134.2 million on 38,372,073 shares outstanding, or $3.17 to $3.50 per share.

For the AI catalyst, I based the analysis on the company’s own cited figures, given that the company cited a “$13 billion proposed U.SSSentinelOne, Inc.$21.40+3.08% Valuation Verdict Overvalued For a complete valuation analysis, visit the SentinelOne, Inc. Stock Analysis page.Run Full Analysis. defense budget for AI and autonomous systems, referenced on the Q4 call, plus a $2.7 billion edge-drone market” in its own November 2025 strategy release, which in turn cited MarketsandMarkets industry research. That’s “$16 billion combined. I assumed a conservative 0.15% eventual capture, ramping to about $24 million of incremental revenue per year by fiscal 2031,” which converted to free cash flow at 18%. I’d rather anchor that conversion rate than simply assert it, given that Diodes converts roughly 9.4% of revenue to free cash flow, while Everspin Technologies (MRAM), a much more comparable memory company that’s still investing heavily in its emerging technology (like GSIT is with Plato), is actually at -14.6%. Meanwhile, Lattice Semiconductor (LSCC), a fully scaled low-power specialty chip company, has seen conversion rates of 19%-34% in recent years on a similar scale. 18% would be slightly better than Diodes’ current performance but well below where Lattice is at scale, which is a reasonable if not spectacular target to aim for given the potential of GSIT’s APU. That puts it at “$4.32 million of mature annual free cash flow. Discounted at a 13%-19% WACC (deliberately high, given single-product, single-technology risk) and haircut 25%-45% for probability of success – zero disclosed APU revenue today, plus the credibility overhang above, that’s a catalyst value of roughly $0.07 to “$0.28 a share, midpoint around $0.14.

Add them up: fair value of $3.25 to $ 3.80 a share, a midpoint of $3.50, against a current price of $ 5.63, so a potential downside of 32 to 42%.

My favorite napkin math was to take my base-business value and subtract it from the current price, which suggests that the market is including a value of about $2.33 a share for the AI option. My own probability-weighted estimate of that option is around $0.14 a share. That’s roughly a 16-times gap. which I believe is a stretch given the lack of APU revenue, silicon more than a year away, and an ongoing fraud investigation into the very benchmark numbers underpinning the story.

For reference on GSIT’s position relative to its peers, here’s operating margin, calculated the same way for all names – TTM operating income divided by TTM revenue, using those same figures from the Seeking Alpha comparisons that were linked:

That’s a -69.6% operating margin for GSIT, which is actually worse than fellow “loss makers” Everspin and AOSL, and substantially lower than the three operating-profitable peers in the comparison universe: Rambus, Netlist, and Diodes. Yet GSIT’s 6.92x EV/sales multiple is actually higher than every one of those profitable peers. The only company in this peer group that has a higher EV/sales multiple than GSIT is Rambus, which is the only peer earning a GAAP operating profit. I’m not comparing GSIT to a company trading at 70x forward earnings. I’m comparing it to the worst operating margin in its own peer group that still trades above every profitable name in that group.

The Best Argument Against Me: Net Cash Is 27% of the Market Cap

I would like to address the other side of the argument, since I think it has some validity. GSIT’s net cash of $58.75 million equals 27.2% of its $216 million market cap is the best ratio in its peer group, versus cash-rich Rambus at 13.6%. No bank debt, and at the guided burn rate of about $16 million, this cash can fund almost 4 years of Plato and Gemini-II development, without another offering. This is a floor, and the reason why I put this issue at “Sell”, versus “Strong Sell”. If you believe the AI story has even modest odds of working, there’s an argument for holding through the noise on a balance sheet this clean.

I would only note that this floor came at a price, as the shares issued jumped from about 25.6 to 38.4 million in the sixteen months following the October 2025 raise, or about 50% dilution. A large part of that cash buffer has already been paid for by existing shareholders.

Risks to My Thesis

If Plato’s timeline is not met again, or if Gemini-II’s proofs-of-concept do not win any orders, my catalyst assumptions become worthless, and the fair value converges towards the base business case. I’d be watching the July 30 earnings call for any changes to the tape-out timeline.

If the securities investigations produce a filed class-action lawsuit, as opposed to the law firms’ circulars, legal expenses and settlements would reduce the value of the net-cash cushion supporting my valuation. Meanwhile, if the investigations are dropped quietly, I’d be less conservative in my probability discount for the AI catalyst.

GSIT has had a large concentration of its distributor and OEM customers, and a contraction of any one of them would reduce my base-case revenue forecasts, since I’ve budgeted zero growth for FY2027 Q1. If the company’s SRAM sales continue to grow steadily and maintain their healthy margins, I could be too conservative in my SRAM multiple discount, as well. The market may well re-rate GSIT’s legacy business higher on its own dime if defense sales continue to ramp up.

Lastly, if Gemini-II secures any disclosed, revenue-generating contracts before the next earnings report, I am going to have to revise my probability estimates considerably higher, which would decrease the value discount due to these concerns substantially.

Bottom Line

The critical question in this one case is $58.75 million – GSIT’s net cash position, and the best cushion in its comp set. But the number that drives my rating is $2.33 versus $0.14: what the market is currently paying for the AI option versus my very conservative estimate of the value today. Until Gemini-II or Plato begins generating booked revenue rather than proof-of-concepts, SBIR phases, and municipal pilots, I’m going to have to stick to my “sell” rating and my fair value estimate of $3.25 to $3.80 per share. And that re-rating is going to happen either after the July 30 print comes out, or after the securities litigation finally gets resolved.