beeese ／咸書

$MRAM $INTC $TSM $GFS $AMD This stock looks massively undervalued when you consider the full picture: highly compelling technology, a strong patent portfolio, and an impressive list of customers and partners. These include IBM, GlobalFoundries, Supermicro, Mitsubishi Electric, TSMC, and NXP Semiconductors. Looking deeper, one of the biggest strengths is the management team, arguably among the most experienced out there. Many come from top-tier companies like Intel, AMD, and Marvell Technology. In my view, people don’t leave large, well-paying industry leaders for a much smaller, lesser-known company without a strong belief in its potential. At the same time, it likely means they maintain valuable connections across the industry, including with major players like Intel and AMD.

$MRAM is one of the most interesting technology monopolies on the market. No position. Not planning one until the execution matches the monopoly. Everspin Technologies is the world’s only manufacturer of Toggle MRAM at scale. Memory that stores data using magnetic states instead of electric charge. Retains information when power goes out. Immune to radiation. Works in extreme temperatures. The US military runs it in missiles, submarines, aerospace systems, and classified platforms. No substitute exists. No competitor operates at this level. The defense contract they just signed has the most interesting clause in any contract. If Everspin ever decides to stop making MRAM the US government requires them to hand over all process technology, IP, and manufacturing know-how to ensure continuity of domestic supply. The Pentagon literally built a legal mechanism to keep this technology alive because they cannot afford for it to disappear. That is not a customer. That is a national security dependency. $40 million contract with the US Department of War just signed. New foundry partnership with Microchip for domestic production in Oregon. The US defense industrial base is onshoring MRAM supply and Everspin is the only game in town. Now the problem. The execution has been frustrating. Revenue growth has been inconsistent. Margins have been all over the place. For a company with this level of monopoly positioning the financial results should look much better than they do. The technology and the contracts are real. The execution is not where it needs to be. $MRAM

$SIVE $LWLG $POET The AI infrastructure supply chain is evolving into a tightly integrated photonics ecosystem, and the strategic alignment between Sivers Semiconductors, Lightwave Logic, POET Technologies, and SilTerra may become one of the most important architectures behind next-generation AI data centers. Here’s how the stack comes together: The Triple-Play Solution This alliance combines three critical technologies into a single photonics platform: • Sivers → InP Continuous Wave Lasers (the light source) • Lightwave Logic → Ultra-fast Polymer Modulators (the data encoder) • POET → Optical Interposer™ integration platform • SilTerra → 8-inch Silicon Photonics manufacturing backbone Together, they address the biggest bottlenecks in AI infrastructure: power consumption, heat, bandwidth density, and scaling limitations. The flow looks like this: Sivers Lasers + LWLG Polymer Modulators → Integrated by POET on SilTerra silicon wafers → Assembled by OSAT/module makers → Deployed into Hyperscaler AI clusters Why the Sivers + LWLG Combination Matters This is the critical technological pairing: • Sivers provides the continuous laser light • LWLG polymers modulate that light into ultra-high-speed data signals At 1.6T and eventually 3.2T interconnect speeds, traditional silicon modulators begin to suffer from thermal and power limitations. LWLG’s polymer modulators are designed to remain highly efficient at those speeds, while Sivers’ lasers provide the stable external light source architecture required for future Optical I/O systems. This is one of the core reasons why External Laser Source (ELS) and Co-Packaged Optics (CPO) are becoming essential for AI infrastructure. SilTerra: The Manufacturing Backbone SilTerra is effectively the manufacturing hub of the ecosystem. The foundry has already integrated LWLG’s polymer process into its silicon photonics PDK, enabling scalable manufacturing of next-generation optical engines on 8-inch wafers. Sivers laser chips are then bonded onto these wafers for POET integration. This transforms the concept from a laboratory demonstration into a manufacturable supply chain. The Final Destination: Hyperscalers The end customers are the companies building the world’s largest AI compute clusters: • $NVDA, future GPU clusters beyond Blackwell will require optical I/O as copper interconnects hit physical limits. • $MSFT and $META both are actively exploring External Light Source architectures to reduce thermal load and improve scalability. • $GOOG heavily focused on Co-Packaged Optics to reduce the massive power consumption of AI training infrastructure. • $AMZN AWS expected to pursue similar photonic scaling architectures for hyperscale AI networking. Current Status (May 2026) The ecosystem has moved beyond the research phase. • SilTerra has reportedly completed initial production wafer runs containing LWLG modulators and POET optical designs. • Alpha-level samples are believed to be under validation with multiple hyperscalers. • Commercial ramp expectations are increasingly centered around late 2026 as AI networking transitions toward optical architectures. The photonics race is no longer theoretical. It is becoming the foundational infrastructure layer of the AI economy.