Welcome to the Rocket Lab team, Mynaric! Today we officially acquired Mynaric, adding laser optical communications to our growing space systems portfolio. A big moment for our teams and for the space industry as we make leading satellite laser communication technology available at the volume and speed demanded by commercial and government satellite customers across Europe, the U.S., and rest of world.

$ARM everywhere ⌁ DD drop 🔖🛸 They said they would do it, they're actually doing it. $ARM is now a direct silicon provider in the agentic AI era. ⌁ Big announcement In agentic AI systems, the CPU serves as the orchestration layer, managing parallel reasoning loops, dynamic tool calling, memory management, accelerator scheduling, data movement, and continuous agent execution. Agentic AI is projected to require >4x current CPU capacity per GW because agents generate far more tokens and inter-agent traffic than static models. This is in this environment that $ARM decided to drop its first-ever production CPU built on Neoverse V3 cores: the Arm AGI CPU, which is ruthlessly optimized specifically for those agentic workloads. The company claims it delivers over 2x performance per rack versus latest x86 on industry workloads while slashing CAPEX by roughly $10b per GW of data-center capacity. This is a big deal. After more than 35 years, $ARM is finally shipping finished chips of its own, not just CPU subsystems. ⌁ Business model leveled up For years $ARM lived on licensing plus royalties. Now they capture direct silicon margin on top of that flywheel. $META is lead co-developer and first customer of the AGI CPU (multi-generation roadmap committed). Other launch partners/customers such as OpenAI, Cerebras, $NET, F5, Positron, Rebellions, SAP, SK Telecom have also been hilighted. CEO Rene Haas clearly laid out the 2031 target: $25b annual revenue and $9 EPS, with the AGI CPU line positioned as a material piece of that expansion. ⌁ Last thoughts I've followed the company for a while now, and this is the clearest execution signal yet under Rene Haas. $ARM is engineering its own high-margin, high-volume product revenue flywheel atop the existing base. If the first racks validate the claims in H2 2026, the revenue path to that $25b target starts looking realistic.

Aeluma $ALMU ⌁ Competitive landscape DD drop 🔖 AI interconnects (CPO/LPO for scale-up/scale-out in GPU clusters) are $ALMU's dominant near-term opportunity (~4B SiPh market 2026 growing 25-45% CAGR to $8-13B by 2030). Co-packaged optics will obviously replace copper. The problem is the optical industry is currently trapped in a massive supply chain constraint. The legacy approach is a dead end for hyperscale. It relies on fabricating InP lasers on tiny three-inch to four-inch wafers, then mechanically bonding them to silicon. ⌁ Aeluma enables scalability The ultimate state of physics here is monolithic direct growth, meaning growing III-V compound semiconductors directly on massive 300mm silicon wafers. $ALMU is taking this exact path. They use proprietary buffer layers and quantum dot active regions to absorb the lattice mismatch between InP and silicon. Since quantum dots are isolated 3D nanostructures, threading dislocations isolate locally instead of killing the whole laser. This allows $ALMU to tap into standard foundry economics. $TSM or $TSEM can process these 300mm wafers. That drops the cost floor by an order of magnitude, enabling high-volume manufacturing. ⌁ Competition Quite frankly, the competitive field is divided between companies solving for tomorrow and companies solving for 2028. $INTC is the big dog in high-volume silicon photonics, currently shipping millions of hybrid transceivers. They don't grow InP directly on silicon like $ALMU. Instead, they pre-fabricate InP dies on native substrates and bond them to silicon waveguides using molecular bonding. It works for low volumes, but structurally fails when AI switches require thousands of integrated lasers per rack. $AVGO and $MRVL dominate the DSP and switch ASIC layer, aggressively pushing Co-Packaged Optics (CPO) using similar bonded or pluggable architectures. Then we have the optical I/O players like Ayar Labs. They bypass the integration physics completely by using an external laser source connected via fiber. They've a massive ecosystem lock-in with $INTC and $NVDA. It gets products to market faster, but it also adds packaging complexity and coupling losses. Incumbents win on execution/relationships today, while $ALMU wins on fundamental scalability/physics (large-wafer monolithic III-V/Si) for cost/performance at AI volumes (millions of lasers/SOAs). Imo, the sharpest technical threats to $ALMU are actually private players. Quintessent is the closest peer. They develop quantum dot lasers and integrate them onto commercial silicon photonics platforms with $TSEM. They focus heavily on high-reliability AI interconnects and have earlier foundry traction. $ALMU counters with larger 300 mm wafers, broader monolithic claims (direct growth vs. hetero bonding), in-house control, and diversified revenue (sensing/quantum de-risks). There is also Ranovus, which is pushing multi-wavelength comb lasers and has strong ties with Cerebras. They lean heavily on hybrid elements for system integration though. ⌁ Last thoughts: why Aeluma $ALMU has the strongest long-term tech moat (true monolithic on 300 mm wafer leading to unmatched scalability/cost at AI volumes) and a confortable $38.6 M cash runway. Their large-wafer direct growth approach also works incredibly well for short-wave infrared sensing too, which enables high-volume defense/mobile applications (U.S. gov diversification: NASA quantum, RFSUNY lasers) and a diversified revenue stream. If they successfully master high-yield, monolithic direct-growth of InGaAs/InP on 300mm silicon at scale, the industry will be forced to license their IP or adopt their wafers. PS: I'll talk about $POET in a dedicated post

$ASML ⌁ NOBODY CAN REPLICATE THIS STACK RIGHT NOW dd drop 🛸🔖 $ASML's state-of-the-art EUV systems are the TWINSCAN EXE:5200B (High-NA) and NXE:3800E (Low-NA) platforms. The physics foundation is identical across both families and rests on 3 non-negotiable pillars that no other wavelength or technology can replicate at scale today. ⌁ First, EUV photons are generated using laser-produced plasma (LPP) at a wavelength of 13.5 nm. Molten tin droplets (≈25 μm diameter, traveling at 70 m/s) are injected into the source chamber 50 000 times per second. A pre-pulse laser flattens each droplet into a thin disk. Then, a main CO₂ laser pulse (10.6 μm wavelength, tens of kilowatts peak power) vaporizes and ionizes the tin, creating a plasma with electron temperatures of tens of eV. The EUV emission arises from 4d–4f electronic transitions in highly ionized Sn atoms/ions; the spectrum peaks sharply at 13.5 nm, the spot where tin conversion efficiency, mirror reflectivity, and minimal absorption all peak. They're demonstrating 1000 watts at intermediate focus, running 500-600 watts in production today. Every lost watt caps wafers per hour. ⌁ Second, the entire optical column is vacuum and mirrors only. EUV photons are absorbed within nanometers by any material, air, glass, quartz, even thin films. Transmission optics are impossible. Every surface in the illuminator and projection optics is a Mo/Si multilayer mirror engineered for Bragg reflection at 13.5 nm incidence angles. Reflectivity compounds multiplicatively; 10-12 mirrors in the optical column reduce transmission to a few percent, which is why source power scaling is existential. In High-NA (0.55 vs Low-NA 0.33), the optics become anamorphic: 4x reduction in the scanning (Y) direction but 8x in the cross-scan (X) direction. This keeps incidence angles on the reticle below the critical grazing angle where multilayer reflectivity collapses. ⌁ Third, resolution is locked by the Rayleigh criterion. The printable critical dimension CD equals roughly k1 x lambda/NA, with lambda is fixed at 13.5 nm. Low-NA at 0.33 gives 13 nm features. High-NA at 0.55 drops it to 8 nm. High-NA therefore eliminates one or more multi-patterning steps for sub-2 nm layers, slashing cycle time, defects, and cost. ⌁ Last thoughts No competitor can replicate this stack. The plasma source, multilayer optics, and vacuum mechatronics required >25 years and €6 billion in R&D; Chinese attempts remain prototypes with orders-of-magnitude lower power and yield. Physics dictates that any alternative wavelength (e.g., 6.7 nm) demands entirely new plasma materials, mirrors, and lasers. It'd lead to another decade-plus effort with no guarantee of higher efficiency.