GSI Technology (GSIT) Q4 2026 earnings review
APU Pivot Burns Cash While Legacy SRAM Shows Cracks
GSI Technology's Q4 tells a story of a business in aggressive transition. While full-year revenue grew 22% to $25.1M, the legacy SRAM foundation is showing severe volatility—key customer Cadence Design Systems dropped to zero sales in Q4. Concurrently, R&D expenses are accelerating as the company doubles down on its APU roadmap (Gemini-II and PLATO). Operating losses widened significantly to $5.2M from $2.3M a year ago. A massive cash infusion in Q3 ($46.9M net) provides the runway needed to survive this transition, but investors must recognize that the company's valuation relies entirely on converting defense and edge AI proof-of-concepts into commercial revenue by late 2027.
🐂 Bull Case
Gemini-II's compute-in-memory architecture boasts a proven performance-per-watt edge, achieving a 3-second Time-to-First-Token at just 30 watts. This makes it highly attractive for power-constrained applications like drones, unmanned vehicles, and smart cities.
The company holds $67.2M in cash with zero debt. At a projected burn rate of ~$4M per quarter, this runway easily extends past the crucial PLATO chip tape-out targeted for early calendar 2027.
🐻 Bear Case
The legacy SRAM business is highly volatile. Cadence Design Systems generated $1.5M in Q1 but zero in Q4. If core revenue continues to erode, it will accelerate the cash burn before the APU products are commercially viable.
The APU segment is still entirely in the Proof-of-Concept (POC) and SBIR (grant) phase. Meaningful commercial design wins and production revenue remain a late-2027 story, representing massive execution risk.
⚖️ Verdict: 🔴
Bearish in the near-term. While the tech validations and government traction are genuinely impressive, the disappearance of a key SRAM customer and the ballooning operating losses highlight the fragility of the bridge connecting today's business to tomorrow's AI promises.
Key Themes
Defense & Edge AI Strategy Gaining Traction
Management is explicitly avoiding hyperscaler data center battles to focus entirely on power-constrained edge applications—a smart macro positioning. This shift is bearing fruit with the U.S. Army SBIR Phase II advancement ($2M) for ruggedized edge AI, and the completion of software for the 'Sentinel' drone POC with G2 Tech. Non-dilutive government funding ($5M to date) validates the tech and offsets R&D burn.
PLATO Development Engine
GSI is heavily investing in PLATO, its next-generation architecture. R&D spending surged from $16.0M in FY25 to $19.9M in FY26 specifically to launch the PLATO chip design. Management highlighted a 'repeatable development platform' that reuses software and system architecture from Gemini-II, drastically lowering future incremental development costs. Tape-out remains scheduled for early 2027.
Extreme Customer Volatility in Legacy SRAM
Management claims the SRAM business provides a 'stable financial foundation', but the data contradicts this narrative. Cadence Design Systems, which accounted for $1.5M (23.9% of revenue) in Q1, plummeted to $0 in Q4. Conversely, KYEC dropped to $267K in Q1 but rebounded to $1.4M in Q4. This wild quarter-to-quarter concentration swing makes forecasting nearly impossible.
Margin Compression Reversing the Trend
After strong gross margins in the first half of the year (58.1% in Q1, 54.8% in Q2), margins have decelerated to 52.4% in Q4. Management attributed this to 'product mix'. As higher-margin SigmaQuad shipments declined from 62.5% of total shipments in Q1 to just 32.6% in Q4, profitability is tracking the wrong direction.
Other KPIs
Accelerating cash burn. Cash used in operations widened from -$10.8M in FY25 to -$17.5M in FY26. Management notes expected burn will normalize to ~$4M per quarter going forward, implying total yearly consumption around $16M. With $67.2M in the bank, the liquidity profile is secure but completely reliant on the October 2025 equity raise.
R&D continues to be the massive weight on the income statement, jumping 87% YoY from $3.0M in 25Q4. While offset slightly by $48K in SBIR funding this quarter, the core engineering costs to execute PLATO are fixed and heavy.
Guidance
Stable. The $6.3M midpoint implies practically zero sequential growth versus 26Q4 ($6.3M) and flat performance YoY compared to 26Q1 ($6.28M). This suggests the SRAM business has plateaued while waiting for the APU products to commercialize.
Accelerating slightly. The 55% midpoint represents a ~260 bps sequential improvement from the 52.4% printed in Q4, indicating management expects a favorable shift back toward higher-margin product mix.
Key Questions
Cadence Design Systems Zero-Out
Revenue from Cadence went from $1.5 million in Q1 to absolute zero in Q4. Is this purely an inventory absorption cycle, a shift in their platform architecture away from your SRAM, or a permanent loss of the account?
Bridging POC to Production
You highlight the Smart City Phase 1 and Sentinel drone POCs. What is the explicit timeline and unit-volume threshold required for these to shift from non-dilutive R&D funding (SBIRs) to recurring commercial purchase orders?
PLATO Tape-Out Costs
You noted cash burn of ~$4M per quarter. Does this run-rate fully bake in the expensive mask sets and tape-out fees required for PLATO at TSMC in early 2027, or should we expect a significant cash utilization spike as that date approaches?