Field Watch — Hardware, AI & Defense Technology, Tracked Daily
Daily brief — Tuesday, November 3, 2026
Hardware & AI
Edge AI · Wireless Systems-on-Chip
Nordic folds Wi-Fi, multiple radios and optional neural acceleration into one battery-focused chip family
Announced November 3, 2026 · Nordic Semiconductor announcement, reported by Unite.AI
Nordic Semiconductor announced its nRF71 Series on November 3, combining Wi-Fi, Bluetooth LE, Thread, application processing and on-device AI support in low-power chips for battery-operated devices. The first parts are the nRF7120, with dual-band Wi-Fi 6, and the nRF7120E, which adds the 6 GHz band for tri-band Wi-Fi 6E and integrates the company's Axon neural processing unit; both also carry a 256 MHz Arm Cortex-M33 application processor and a 256 MHz RISC-V coprocessor. Nordic states standby power up to 20 times lower than its current nRF70 Series Wi-Fi products and says the Axon unit runs machine-learning models 10 to 17 times faster than the Cortex-M33 alone (both vendor-stated). Unlike the earlier companion-chip line, the new parts include the host processor, so a design no longer needs a separate microcontroller, and the series is sampling with selected customers on a project basis rather than in general availability.
Why it matters: For battery devices, the win is integration — one chip carrying connectivity, the application and optional inference removes parts and board area while keeping standby draw low enough for Wi-Fi to stay connected.
Read the source: Nordic Semiconductor combines Wi-Fi and edge AI in nRF71 chips (Unite.AI)
AI Accelerators · Inference Systems
Supermicro will build inference systems pairing Nvidia GPUs with SambaNova's SN50 chips
Announced November 3, 2026 · Company announcement via BusinessWire, syndicated
SambaNova announced on November 3 that Supermicro is entering a strategic collaboration, initially planning for 10,000 SN50 chips, to build and sell disaggregated inference systems that combine its Reconfigurable Dataflow Units with GPUs. In the pairing the companies describe, Nvidia GPUs handle the prompt-reading prefill phase while the SN50 chips handle the response-writing decode phase, and Supermicro will integrate, test, deliver and support the systems, including on-site service, with SambaNova engineers helping tune models and deployments. The SN50 is SambaNova's fifth-generation RDU, built for inference, and the companies say they will start with NeoClouds and sovereign clouds before expanding to enterprises running AI in their own data centres. No financial terms for the collaboration were disclosed in the announcement.
Why it matters: Inference is splitting into phases with different hardware economics — a mainstream system builder selling GPU-plus-specialist systems lets buyers add inference capacity inside existing data centres instead of waiting for new builds.
Read the source: SambaNova SN50 chips to power Supermicro's new inference racks (BusinessWire, syndicated)
Memory & Supply Chain
Memory · U.S. Production Options
SK hynix weighs using Intel's Ohio site for DRAM as it maps a wider U.S. memory footprint
Reported November 3, 2026 · Seoul Economic Daily, from industry officials and foreign media reports
SK hynix is considering producing DRAM at the semiconductor plant Intel is building in New Albany, Ohio, according to industry officials cited on November 3, with options under discussion including leasing part of the site or forming a joint venture with Intel and major U.S. cloud companies. The report notes the Ohio site already holds a 2023 Foreign-Trade Zones Board approval covering memory-chip production, and that Intel pushed completion of its first plant there back to 2030, with operations expected between 2030 and 2031. If it proceeded, Ohio would be SK hynix's first front-end DRAM fab in the United States, alongside its Indiana HBM packaging facility targeted for mass production in 2029 and the New York NAND plant its Solidigm subsidiary is pursuing. Analysts cited in the report flag real hurdles — DRAM and logic lines differ in process and equipment, U.S. production costs are high, and the decision is expected to hinge on long-term supply agreements — and an SK hynix official said there was nothing specific the company could confirm.
Why it matters: This is an option being weighed, not a commitment — but a memory maker with output effectively sold out through 2029 looking at an existing U.S. fab shell shows how far buyers' long-term agreements now reach into fab geography.
Read the source: SK hynix weighs DRAM output at Intel's Ohio fab (Seoul Economic Daily)
Memory · HBM Base Dies in Texas
Samsung's second Texas plant is reported set to make HBM base dies for AMD and Nvidia
Reported November 3, 2026 · Reuters, from Maeil Business Newspaper reporting citing industry sources
Samsung Electronics will use its second foundry plant in Taylor, Texas, to produce base dies for high-bandwidth memory chips for AMD and Nvidia, Reuters reported on November 3, citing South Korea's Maeil Business Newspaper and its industry sources. Maeil reported that construction is due to start formally this month, that the plant has secured orders for its entire output with most of the secured volume for AMD, and that the facility will use 2-nanometre technology across about 2.7 million square feet. Equipment installation is expected from February 2029, with mass production in 2030, according to the same reporting. Samsung declined to comment, so the customer and volume details remain source-reported rather than company-confirmed.
Why it matters: The base die is where HBM meets logic — placing that work in Texas on a 2-nanometre line would put a critical, custom part of the AI memory stack inside the U.S. for two of its largest buyers.
Read the source: Samsung to make AMD, Nvidia chip parts at second Texas plant, Maeil Business reports (Reuters, syndicated)
Defense & Industry Watch
Defense · Air Defence Exports
ASELSAN signs a $1.1 billion air-defence export deal with the buyer undisclosed
Announced November 3, 2026 · Company announcement, reported by Defensehere
ASELSAN announced on November 3 that it has signed an export contract valued at $1.1 billion with an international customer for the sale of air defence systems. Neither the customer nor the specific systems to be sold were disclosed in the announcement coverage. The company's air-defence portfolio spans its layered Steel Dome architecture, including the SIPER long-range system, the HISAR family for medium-to-low altitude defence, the KORKUT 35 mm gun system for short-range threats, and the GURZ hybrid gun-and-missile platform. The announcement follows a separate $125 million export contract the company reported last month for weapons and electronic warfare systems.
Why it matters: A billion-dollar export order with the system mix undisclosed still signals demand depth for layered air defence — the category where missiles, guns and drone defence now have to be bought as one architecture.
Read the source: ASELSAN signs $1.1 billion contract for air defense systems export (Defensehere)
Defense · Financial Systems Modernisation
CACI takes a five-year, $108 million Air Force award to sustain and modernise its accounting system
Announced November 3, 2026 · Company announcement via PR Newswire
CACI International announced on November 3 that the U.S. Air Force has awarded it a five-year, $108 million contract to sustain and modernise the Defense Enterprise Accounting and Management System, known as DEAMS. The company said the work applies Agile delivery, automated testing, cybersecurity and cloud-enabled operations to streamline financial operations and improve visibility across the enterprise. The announcement builds on DEAMS having recently received an unqualified audit opinion, and frames the modernisation as supporting faster, data-driven decisions for global financial operations. No task-order breakdown or place-of-performance detail was given in the announcement.
Why it matters: Audit-ready financial systems are operational infrastructure for a service — a five-year sustainment award shows the work has moved from achieving a clean audit opinion to keeping it while the system modernises.
Read the source: CACI advances U.S. Air Force financial transparency and mission readiness (PR Newswire)
Work with SALAR
Tracking the field is the job. Acting on it is the engagement.
Field Watch is the public side of the monitoring SALAR Technologies does every day. The paid side is what we do with it for a specific program: BOM provenance and review, supply-chain resilience analysis, hardware and semiconductor design review, and manufacturing-stability work — founder-direct, scoped, and deliverable-first.
See what we can start now: salartechnologies.com/capabilities
Archive
Daily brief — Monday, November 2, 2026
Hardware & AI
Edge AI · Production Silicon
Ceva's neural processor reaches production silicon for the first time inside Mobilint's edge chip
Announced November 2, 2026 · Ceva company announcement, reported by Securities.io
Ceva said on November 2 that Mobilint has deployed its SensPro AI digital signal processor and NeuPro-Nano neural processing unit in REGULUS, Mobilint's standalone edge AI system-on-chip — the first production deployment of Ceva's NeuPro NPU family. Ceva states the chip delivers 10 TOPS of AI performance at under 3 watts and supports up to 8 GB of memory (vendor-stated), splitting work between the SensPro engine for sensing and signal processing and the NeuPro-Nano engine for always-on neural inference. Target uses are power- and space-constrained devices including robotics, drones, machine vision, smart cameras, and IoT equipment. Ceva also said a REGULUS-based product, Mobilint's MLD-R1 AI USB accelerator first previewed at COMPUTEX 2026, is now shipping for adding on-device AI to existing systems.
Why it matters: Licensed neural IP only proves itself in someone else's shipping silicon — a first production deployment, in a sub-3-watt standalone processor rather than a data-centre part, moves the edge-AI contest to complete on-device systems.
Read the source: Mobilint REGULUS marks first production deployment of Ceva NeuPro NPU (Securities.io)
AI Infrastructure · Data Centres
Samsung SDS breaks ground at Gumi on a 60-megawatt AI data centre built to double
Groundbreaking November 2, 2026 · Seoul Economic Daily, from the company's ceremony
Samsung SDS held a groundbreaking ceremony on November 2 for an AI data centre in Gumi, North Gyeongsang Province, on a site of roughly 46,000 square metres. Power capacity is planned to start at 60 megawatts and expand to as much as 120 megawatts, with completion scheduled for the end of 2028. The design uses air cooling for general server areas and water cooling for high-density GPU areas, and the facility is intended to serve both Samsung's own GPU service and colocation customers running their own equipment. The report places the centre inside Samsung's wider robot-transformation work and a group plan to invest a combined 60 trillion won in the Yeongnam region.
Why it matters: The build is phased around power, not floor space — starting at 60 megawatts with a designed path to double is now the standard shape for AI capacity, because energised capacity, not buildings, is the scarce input.
Read the source: Samsung breaks ground on Gumi AI data center (Seoul Economic Daily)
Memory & Supply Chain
Memory · Long-Term Supply Agreements
Four memory makers now hold 58 long-term deals, and five years with a deposit is the reported default term
Reported November 2, 2026 · FourWeekMBA, from Hankyung and Datanews reporting of November 2
Counting the long-term agreements each company has reported, Samsung Electronics, SK hynix, Micron, and Sandisk hold 58 long-term memory agreements between them — 26 at Micron, 14 at SK hynix, 10 signed at Samsung with 10 more in negotiation, and 8 at Sandisk — with the count covering agreements, not unique buyers. Hankyung reported on November 2, citing the industry, that a five-year base term with deposits of 20% to 25% or take-or-pay terms has become the default shape for these deals; the companies have not disclosed deposit amounts for individual contracts. SK hynix's 14 agreements are reported to cover more than 50% of its total production capacity for 2027 and 2028, and about 80% of Samsung's memory output for next year is reported to already have a confirmed buyer.
Why it matters: Memory buying has moved from quarterly pricing to multi-year, deposit-backed capacity claims — a buyer without an agreement is no longer competing for output, only for what the agreements leave behind.
Read the source: Samsung, SK hynix, Micron and Sandisk hold 58 AI memory deals (FourWeekMBA)
Memory · China DRAM Supply
China's CXMT now has conventional DRAM scale, while reported HBM yields still separate it from the AI memory market
Analysis published November 2, 2026 · Sector Foundry, from company results and Reuters reporting
An analysis published on November 2 finds that ChangXin Memory Technologies has reached meaningful scale in conventional DRAM — it reported third-quarter revenue of RMB 118.71 billion on October 30, and Counterpoint Research estimated it held 10% of global DRAM revenue in the second quarter — while advanced AI memory remains a different contest. Reuters reporting cited in the analysis says CXMT supplied HBM3E samples to domestic AI-chip designers and planned limited production, with source-estimated usable yields of approximately 20% to 30% against around 70% for SK hynix; those yield figures are source-reported and not independently audited. Reuters has also reported that CXMT projects in Beijing, Shanghai, and Hefei could eventually double monthly capacity to approximately 600,000 twelve-inch wafers, a potential future scale rather than current finished output.
Why it matters: New conventional supply and scarce HBM can coexist — added Chinese DRAM may pressure standard memory prices through import substitution while leaving accelerator memory allocation, the binding constraint, largely unchanged.
Read the source: Can China's DRAM makers break the global memory cycle? (Sector Foundry)
Defense & Industry Watch
Defense · Directed Energy
Leonardo DRS joins the Army's first laser production programme for power and thermal work
Announced November 2, 2026 · Company announcement via GlobeNewswire
Leonardo DRS announced on November 2 that AeroVironment has selected it to provide rugged integration expertise for the U.S. Army's Enduring-High Energy Laser production programme, applying its power management and environmental control work to a mobile system intended to counter unmanned aircraft. The subcontract supports AeroVironment's production of the LOCUST X3 laser weapon system under the programme; the company describes the X3 as a 30-kilowatt, platform-agnostic system designed for platforms including the Joint Light Tactical Vehicle, with palletised configurations also available. The Army awarded AeroVironment the programme in September 2026, describing it as the service's first production contract for a high-energy laser weapon system. No subcontract value was disclosed in the announcement.
Why it matters: A laser weapon is mostly a power and cooling problem mounted on a vehicle — putting the production subcontract for exactly those subsystems in place is what moves directed energy from demonstrations to a fieldable programme.
Read the source: Leonardo DRS selected for U.S. Army Enduring-High Energy Laser production program (GlobeNewswire)
Daily brief — Sunday, November 1, 2026
Hardware & AI
AI Accelerators · Shipping Specifications
As Rubin systems reach customers, the final shipping figures trim memory and link bandwidth — and the 8-GPU system loses memory capacity
Published November 1, 2026 · ServeTheHome, from NVIDIA's updated architecture datasheet first posted in September
ServeTheHome reports that NVIDIA has locked down the shipping specifications for its Rubin GPU now that the first Vera Rubin NVL72 systems are becoming generally available at hyperscalers, and the final vendor-published figures sit below the preliminary ones in two places: HBM4 memory bandwidth is 19.2 TB/s per GPU, down from a preliminary 22 TB/s (about 9.375 Gb/s per pin instead of 10.75 Gb/s), and NVLink scale-up bandwidth is 3.0 TB/s, down from 3.6 TB/s — both vendor-stated. Memory capacity is unchanged at 288 GB per GPU and the stated compute figures are unchanged, while scale-out networking bandwidth moves the other way, up from 0.4 to 0.45 TB/s. Field Watch noted the per-GPU bandwidth revision on October 28; this report completes the picture at system level, with NVL72 rack totals revised to 1,400 TB/s of memory bandwidth and 216 TB/s of NVLink, and the 8-GPU HGX Rubin NVL8 losing capacity as well — 2.0 TB of HBM4 instead of 2.3 TB, which the report works out to roughly 250 GB per GPU, effectively seven of the eight stacks, with memory bandwidth down from 176 to 130 TB/s.
Why it matters: Preliminary figures are planning inputs until hardware ships — the shipping datasheet, not the launch presentation, is what cluster designs, power budgets, and competitive comparisons should now be built on.
Read the source: NVIDIA finalises Rubin shipping specifications as production hardware arrives (ServeTheHome)
Power Semiconductors · AI Data-Centre Power
Infineon pairs a new German power-semiconductor lab with a 300-millimetre gallium nitride bet in Austria
Published November 1, 2026 · Ad Hoc News roundup of company announcements
Infineon has opened a new office and laboratory building at Warstein in Germany, designed for around 350 employees with roughly 1,200 square metres of laboratory space, alongside its wider power-semiconductor and AI data-centre work. At Villach in Austria, the company plans to invest around EUR 1.1 billion through 2034 to establish what it describes as the world's first high-volume manufacturing of GaN power semiconductors on 300-millimetre wafers (vendor-stated), with production scheduled to begin by the end of 2027 and EUR 319 million in funding earmarked for the project under the European Chips Act, per company statements cited in the report. The build-out follows the October 21 broadening of Infineon's collaboration with NVIDIA on GaN-based power conversion for 800-volt DC architectures in AI data centres. Infineon has provisionally scheduled November 10 for its fourth-quarter and full-year 2026 results.
Why it matters: AI data-centre growth is increasingly a power-delivery problem — capacity in wide-bandgap power devices, and the labs that qualify them, gate how fast new compute can actually be energised.
Read the source: Infineon expands its GaN and AI power work (Ad Hoc News)
Memory & Supply Chain
Memory · U.S. Packaging Operations
SK hynix puts its lead HBM developer in charge of a new Americas division as the Indiana packaging plant advances
Reported November 1, 2026 · East Asia Brief, from Korean reporting; reorganisation effective November 1
SK hynix has named Chief Development Officer Ahn Hyun — the engineer who led development of its fifth-generation HBM3E memory — to run a newly formed Americas division, in an appointment announced on October 30 and effective November 1. The division oversees the company's $4 billion advanced packaging complex in West Lafayette, Indiana, described in the report as its largest capital project outside South Korea, and places senior memory engineering leadership closer to U.S. hyperscale customers. The Indiana lines are to focus on next-generation HBM4 and HBM4E parts built around customised base dies, with the reported schedule calling for cleanroom installation to complete by October 2028 and mass production in the second half of 2029. Cha Sun-yong, head of the Future Technology Research Institute, succeeds Ahn as Chief Development Officer.
Why it matters: Custom HBM is co-designed with the processor it sits beside — putting the developer of the current generation in charge of the U.S. packaging site is an organisational commitment to that co-design happening in North America, years before the plant produces.
Read the source: SK hynix names lead HBM developer to head new Americas unit (East Asia Brief)
Supply Chain · Technology Policy
Quad ministers pledge over $85 million for trusted AI access and task a hunt for semiconductor chokepoints
Announced October 31, 2026 · Quad foreign ministers' joint statement in Atlanta, reported by Reuters
Foreign ministers from the United States, Australia, India, and Japan, meeting in Atlanta on October 31, stated their intent to provide more than $85 million in technical and financial assistance to support access to trusted artificial intelligence across the Indo-Pacific, including through the Quad SI Accelerator; the statement sets no country-by-country allocations or spending timetable. Alongside the funding intent, the four governments said they will launch an effort to identify vulnerabilities and chokepoints in semiconductor and related supply chains before those weaknesses cause disruptions. The joint statement also expressed grave concern about economic coercion through arbitrary export restrictions and non-market practices that disrupt supply chains in critical industries, without naming a country in that passage. It was the third Quad foreign ministers' meeting this year.
Why it matters: The commitment is modest in money and specific in method — mapping semiconductor chokepoints in advance is the same discipline buyers are now applying privately, moved to government level.
Read the source: Quad ministers work on supply chains and trusted AI (Reuters, syndicated)
Defense & Industry Watch
Defense · Satellite Sustainment
Space Force plans a sole-source sustainment award of up to $945.9 million after Boeing declines to sell its satellite data rights
Reported October 30, 2026 · Washington Technology, from the Space Force's October 27 SAM.gov notice
The U.S. Space Force says it has no practical alternative to a sole-source sustainment contract with Boeing worth up to $945.9 million, because Boeing will not sell the proprietary data another contractor would need to do the work. Boeing built the Wideband Global SATCOM and Ultra High Frequency Follow-On satellites and key parts of their ground control system, and the government's October 27 notice says the company has repeatedly declined requests to sell the necessary data rights — data developed at private expense, including detailed design documentation for the Boeing 702HP and 702MP satellite platforms, simulators, and ground control software. The award has not yet been made and a comment period runs until November 17; the anticipated contract carries a five-year base period and a single 10-year option, covering life-cycle engineering, sustainment, cybersecurity, and modernisation across several military satellite communications constellations, including bringing WGS-11, WGS-12, and the Protected Tactical SATCOM Prototype payload into sustainment.
Why it matters: Who owns the design data decides whether sustainment can ever be competed — this case prices that lock-in directly, at up to $945.9 million over as long as fifteen years.
Read the source: Space Force plans sole-source Boeing sustainment award over data rights (Washington Technology)
Defense · Naval Air Defence
BAE's deck-mounted launcher passes its first live firing, aiming to let Navy ships rearm at sea
Announced October 27, 2026 · Company announcement, reported by Defence Blog
BAE Systems has completed the first live-fire test of its Adaptable Deck Launcher at China Lake, California, the company announced on October 27. The launcher sits on a ship's deck in fixed-angle canisters rather than in the deep below-deck wells of the Mk 41 Vertical Launching System, fires Evolved SeaSparrow Missiles from Mk 25 Quad-Pack canisters holding four interceptors each, and reuses the Mk 41's electronics and logistics suite. Because the canisters are above deck, spent ones can be replaced at sea with standard shipboard equipment, so a ship does not have to return to port to rearm. The system was developed under a U.S. Department of War contract for the NATO Seasparrow Project Office, is designed for aircraft carriers, amphibious ships, and other surface combatants with compact configurations for medium unmanned surface vessels, and the company released no details of how many missiles were fired or which ESSM version was used.
Why it matters: Magazine depth and rearming are the pacing problems for naval air defence — a launcher that adds interceptors to ships built without vertical launch wells, and reloads underway, attacks both without new hull construction.
Read the source: BAE Systems test-fires new deck launcher for U.S. Navy (Defence Blog)
Daily brief — Saturday, October 31, 2026
Hardware & AI
Semiconductors · Quarterly Results
Intel's data-centre business jumps while its PC business shrinks under memory pressure
Reported October 31, 2026 · Seoul Economic Daily, from Intel's third-quarter results released October 29
Intel's third-quarter revenue rose 21% from a year earlier to $16.51 billion, with earnings per share of 52 cents against a 41-cent estimate — company-reported figures carried in the coverage. The growth came from servers: data centre and AI revenue reached $7.3 billion, up 78%, and foundry revenue was $6.7 billion, up 58%. The client computing division, which makes PC chips, fell 5% to $8.1 billion, and the report says that unit took the full brunt of surging PC memory prices as major memory makers concentrate production on high-bandwidth memory for AI. Intel guided fourth-quarter revenue to $15 billion to $16 billion, below the $17.1 billion LSEG estimate, citing a PC market downturn it expects to persist through the first half of next year, and its shares fell more than 30% intraday on October 30.
Why it matters: The same memory reallocation that powers AI servers is now visibly shrinking the PC side of a major chipmaker's business — memory economics are setting the shape of the whole processor market, not just accelerator supply.
Read the source: Intel plunges 30% on PC worries (Seoul Economic Daily)
AI Infrastructure · Server CPUs
Independent tests put Nvidia's Vera server CPU ahead on memory bandwidth and Python, behind on cryptography and compression
Reported October 30, 2026 · BigGo Finance, summarising Tom's Hardware testing
Independent testing of Nvidia's 88-core Vera processor, built on its custom Olympus core, found it matching or beating comparable AMD EPYC Turin and Intel Granite Rapids-AP systems across a range of general-purpose server workloads, according to Tom's Hardware testing summarised on October 30. Each Vera socket pairs 88 cores and 176 threads with up to 1.5 TB of SOCAMM2 LPDDR5X memory for roughly 1.2 TB/s of aggregate bandwidth. On SPEC CPU 2026 intrate, a dual-socket Vera system edged a dual EPYC 9755 system by about 1.5% and finished 8.6% ahead of a dual Xeon 6980P on generic GCC builds, and it led the PyBench average by 4% over the EPYC 9755, with a clear lead in the STREAM memory-bandwidth benchmark — all figures are reported test results, not vendor-stated. The results were not a clean sweep: Vera trailed in OpenSSL, Zstd decompression, and 7-Zip work, while separate Phoronix testing found clean scaling from one socket to two.
Why it matters: Agentic AI workloads run large amounts of ordinary code around the accelerators — a credible standalone CPU from the GPU leader, strongest exactly in memory bandwidth and Python, changes the host-processor choice for AI servers from a default to a decision.
Read the source: Nvidia's Vera CPU takes on Intel and AMD in server benchmarks (BigGo Finance)
Memory & Supply Chain
Memory · Supplier Coordination
SK hynix takes its supplier Partners Day to Japan for the first time to align on AI memory
Reported October 30, 2026 · Aju Press, from SK hynix's announcement
SK hynix brought together executives from more than 20 Japanese semiconductor suppliers in Chiba Prefecture on October 29 for its first Partners Day held in Japan, the company said on October 30. The invited materials, equipment, and testing companies included Shin-Etsu, SUMCO, JSR, Tokyo Electron, Disco, and Advantest. SK hynix shared its medium- to long-term business strategy and technology roadmap so partners can anticipate future demand and prepare investments and production capacity as the AI memory market expands. The gathering did not announce a new manufacturing investment: the company said it is examining options to strengthen its memory business, including additional production bases, but that no decision has been finalised. It comes as the company accelerates advanced high-bandwidth memory production after reporting record third-quarter operating profit of 83.33 trillion won earlier in the week.
Why it matters: HBM output depends on Japanese wafers, chemicals, grinding, and test capacity as much as on SK hynix's own fabs — aligning those suppliers on the roadmap years ahead is how a sold-out memory maker protects its ramp.
Read the source: SK hynix deepens Japan chip supply ties (Aju Press)
Defense & Industry Watch
Defense · Air and Missile Defence
United States clears a possible $2.1 billion sale of 500 Patriot interceptors to Ukraine, paid for by allies
Approved October 30, 2026 · U.S. State Department notification, reported by Defence Blog
The U.S. State Department approved a possible sale of 500 PAC-3 MSE interceptors for Ukraine's Patriot air defence systems on October 30 and notified Congress the same day, with the package estimated at $2.1 billion. The department said the purchase will be paid for entirely by partners and allies: donations from Canada, Germany, and Norway through the JUMPSTART mechanism, plus Ukrainian national funds financed by the European Union's Ukraine Support Loan, with no U.S. funds involved. The package also includes classified and unclassified spare parts and logistics and program support, and Lockheed Martin is the principal contractor. The notification is not a contract — it opens a congressional review period before a letter of offer and acceptance can be signed and an order placed — and the State Department gave no delivery schedule.
Why it matters: For Patriot, interceptor supply has been the tighter constraint than launchers — this approval tests whether allied-funded orders can move through a production queue that is already committed to several other buyers.
Read the source: U.S. clears sale of 500 Patriot PAC-3 missiles to Ukraine (Defence Blog)
Defense · Space Systems
ESI Motion to supply solar-array drive electronics for Lockheed Martin's Space Development Agency work
Announced October 30, 2026 · Company announcement via SpaceNews
ESI Motion announced on October 30 that it is supplying its Solar Array Drive Electronics to Lockheed Martin for Lockheed Martin's work on the Space Development Agency's Proliferated Warfighter Space Architecture programmes. The drive electronics consist of the company's space-rated, flight-heritage SuperNova and Pulsar motor controllers, which manage the solar arrays that power the satellites. The architecture is described as a network of satellites providing secure communication, missile tracking, and space-based support. The announcement did not disclose a contract value, quantities, or a delivery schedule.
Why it matters: A proliferated constellation lives or dies on repeatable subsystem supply — solar-array drive electronics are a small, mission-critical part that has to be delivered consistently across many spacecraft, not once for a single satellite.
Read the source: ESI supplying solar array components to missile defense programs (SpaceNews)
Daily brief — Friday, October 30, 2026
Hardware & AI
AI Accelerators · Custom ASICs
MediaTek adds a second custom AI chip customer as its first accelerator starts shipping
Reported October 30, 2026 · Third-quarter earnings call, reported by TrendForce
MediaTek confirmed on its third-quarter earnings call on October 30 that it has a second customer for its custom data-centre AI accelerator chips, with co-design work already underway and mass production expected to begin in 2028. The company said the new customer is not a Chinese company, and it did not disclose the customer's identity, the chip specifications, or expected revenue from the programme. Its first AI accelerator is further along, having begun small-volume shipments in September, and the company said demand for it remains strong. MediaTek also said several dies in its second-generation multi-die ASIC design have already taped out, with the larger and more complex dies still on the original timeline, and that it is working with a U.S. integrated device manufacturer and substrate suppliers on advanced-packaging yields. Third-quarter consolidated revenue was NT$168.06 billion, up 18.3% from a year earlier, with gross margin at 46.3% — both company-reported figures.
Why it matters: Custom accelerators are no longer a one-customer business for MediaTek — a second programme reaching production in 2028 widens the non-GPU route hyperscale buyers can plan around, provided packaging yields keep pace.
Read the source: MediaTek confirms second AI ASIC customer for 2028 production (TrendForce)
Memory & Supply Chain
Memory · Long-Term Supply Agreements
Samsung puts more of its memory output under long-term deals as it negotiates with more buyers
Reported October 30, 2026 · SemiMedia, from the company's October 29 earnings call
Samsung's memory business has signed multiyear supply agreements with 10 major customers and is negotiating with another 10, and it aims to place 60 to 70 percent of its memory production capacity under long-term agreements by 2030. In the third quarter, HBM4 revenue more than tripled from the previous quarter and represented about 60 percent of Samsung's total HBM revenue. The company expects DRAM supply to remain tight in 2028, noting that higher investment does not translate immediately into additional output because new factories and equipment take time to enter production. Third-quarter DRAM and NAND shipments each rose by a single-digit percentage from the previous quarter, while average selling prices rose about 20 percent and nearly 10 percent respectively — all company-reported figures from the call coverage. Separately, Samsung's foundry business plans to increase investment in South Korea and the United States and is considering additional wafer fabrication and advanced packaging capacity.
Why it matters: The market is shifting from spot buying to multi-year capacity reservations — a buyer without a long-term agreement is increasingly competing for the remainder, not for the full output.
Read the source: Samsung says 2027 HBM supply is sold out (SemiMedia)
Supply Chain · PC Manufacturing
Laptop makers move some production back toward China as memory drives up the bill of materials
Reported October 29, 2026 · Money Today, citing Nikkei reporting and TrendForce data
HP, ASUS, and Acer are moving part of their PC production volumes from Thailand and Vietnam back to China, and Dell is considering a similar plan, according to Nikkei reporting cited by Money Today on October 29. The shift is a partial reversal of the diversification strategy that spread notebook production outside China, and the report attributes it to cost pressure as memory and storage prices have risen. TrendForce data cited in the report puts the combined bill of materials for CPUs, DRAM, and SSDs in a $900-class laptop at approximately 68% in the third quarter of this year. The same data projects the share of notebook production outside China falling from about 24% last year to 21% this year, and potentially below 20% next year. Access to the supply chain for Chinese-made memory is also cited as a competitive factor for local contract manufacturers.
Why it matters: When processors, memory, and storage take up more than two-thirds of a laptop's component cost, assembly-location decisions follow component economics — supply-chain geography is being repriced by the memory shortage, not only by tariffs or policy.
Read the source: China bets on PC production and HBM despite low yields as Samsung and SK hynix boom (Money Today)
Defense & Industry Watch
Defense · Counter-Drone Systems
Two U.S. firms pair an autonomous weapon station with RF sensing for layered counter-drone defence
Announced October 30, 2026 · Company announcement via GlobeNewswire
Allen Control Systems and CX2 announced a strategic partnership on October 30 to advance integrated counter-unmanned aircraft system capabilities for the United States and its allies, covering mobile, expeditionary, fixed-site, and maritime operations. The work follows the inclusion of both companies' technologies in the Joint Interagency Task Force 401 counter-drone marketplace, a Department of Defense initiative intended to accelerate procurement and deployment of such capabilities. The pairing combines ACS's Bullfrog autonomous weapon station, which uses computer vision, autonomous control, and precision robotics to turn existing weapon systems into counter-drone solutions, with CX2's electronic warfare and radio-frequency sensing systems, which detect, classify, and geolocate RF emissions to identify threats and enable electronic disruption. Both companies emphasised open architectures and rapid integration, and framed the effort as layered defence that does not rely exclusively on expensive interceptors.
Why it matters: Against low-cost drones, the economics only work if sensing, electronic disruption, and a low-cost kinetic effect arrive as one integrated system — this partnership is explicitly built around that cost exchange.
Read the source: ACS and CX2 partner to advance integrated counter-drone defense capabilities (GlobeNewswire)
Defense · Reconnaissance Satellites
Hanwha Systems wins South Korea's small radar-satellite constellation work on a sharply lower bid
Reported October 30, 2026 · Seoul Economic Daily, from the KONEPS bid opening on October 30
Hanwha Systems ranked first in the bid opening on October 30 for a contract from South Korea's Agency for Defense Development to develop a constellation of small synthetic aperture radar satellites, according to the Public Procurement Service's KONEPS platform as reported by Seoul Economic Daily. Hanwha bid 368.47 billion won against a base price of 567.2 billion won, while rival Korea Aerospace Industries bid about 460.3 billion won — a gap of 91.8 billion won between the two offers. The programme places dozens of reconnaissance satellites of about 100 kilograms each in low Earth orbit to monitor the Korean Peninsula and surrounding waters in near real time, with radar imaging that works at night and in bad weather; the original goal was more than 40 small satellites by 2030, and this contract runs through November 2031. Hanwha has developed the SAR sensor and data link for the military's Project 425 reconnaissance satellites since 2018, and its Jeju Space Center is equipped to assemble, test, and produce up to 100 satellites a year, per the report.
Why it matters: A constellation award decided in part on price, to a bidder with an existing mass-production line, shows reconnaissance constellations are now being bought as a production programme — revisit rate depends on building many satellites consistently, not on a single exquisite spacecraft.
Read the source: Hanwha Systems wins military satellite deal (Seoul Economic Daily)
Daily brief — Thursday, October 29, 2026
Hardware & AI
Programmable Logic · FPGA Portfolio
Altera lays out a four-family FPGA line built around AI systems, led by Agilex 27
Announced October 28, 2026 · Company announcement via BusinessWire
Altera announced Agilex 2X, a new FPGA and SoC portfolio spanning four families aimed at different points on the performance-to-power spectrum. Agilex 27, the first family to launch, targets data-centre AI infrastructure, terabit networking, 6G and satellite networks, and aerospace and defence work, with connectivity the company lists up to 256 Gbps transceivers and hardened support for 800GbE and 1.6TbE Ethernet, PCIe 6.0 and 7.0, and DDR5 and LPDDR memory interfaces. Altera states the new fabric delivers 1.8 times the performance per watt of the previous Agilex 7 generation (vendor-stated). The other families cover video and sensors (Agilex 25), robotics and industrial automation (Agilex 23), and board control and interface work (Agilex 22). Quartus Prime support for Agilex 27 reaches early-access customers in mid-November, and engineering sample silicon is expected in the second half of 2027.
Why it matters: A single programmable platform that scales from data-centre datapaths down to edge control — with samples still a year out — is a long-lead design decision: teams choosing it now are committing to a roadmap, not to silicon in hand.
Read the source: Altera announces Agilex 2X FPGA and SoC portfolio (BusinessWire, syndicated)
AI Infrastructure · Optical Interconnect
STMicro raises its AI data-centre outlook as optical links drive demand
Reported October 29, 2026 · Reuters, from the company's third-quarter results
STMicroelectronics raised its 2026 forecast for AI data-centre revenue to about $1.3 billion, up from a July expectation of more than $1 billion, and kept its 2027 goal of well above $2 billion — both figures are company forecasts, not booked sales. The company said demand was driven mainly by its silicon photonics chips, which move data between servers as light rather than electrical signals. Third-quarter revenue was $3.76 billion, and the company forecast fourth-quarter revenue of $4.15 billion, give or take 3.5%. It also said demand from low-Earth-orbit satellite makers contributed, alongside a broader recovery in communications equipment and computer peripherals.
Why it matters: The growth is in the links between accelerators, not the accelerators themselves — interconnect and photonics capacity is becoming a planning constraint alongside compute and memory.
Read the source: STMicro lifts data centre revenue forecast after Q3 profit beat (Reuters, syndicated)
Memory & Supply Chain
Memory · HBM Supply
Samsung says next year's HBM is already sold out, with the market tight through 2028
Stated October 29, 2026 · Third-quarter earnings call, reported by Seoul Economic Daily and Reuters
On its third-quarter earnings call on October 29, Samsung Electronics said its high-bandwidth memory supply for 2027 is already sold out, with pricing negotiations completed at target levels, and that major customers continue to ask for additional volumes. Samsung said HBM4 now accounts for about 60% of its HBM revenue, and that early customer evaluations of its HBM4E samples — first shipped in May — have drawn positive feedback. Reuters reported the same day that Samsung expects the chip market to stay tight through 2028 because new fabrication plants take time to build. For comparison, rival SK hynix said a day earlier that its DRAM and HBM chips were sold out through 2029, per the Reuters report.
Why it matters: Two of the three HBM makers now describe supply as committed years ahead — allocation, secured early, remains the difference between a program that ships and one that waits.
Read the sources: Samsung says 2027 HBM supply sold out (Seoul Economic Daily) · Samsung sees AI chip shortage worsening (Reuters, syndicated)
Defense & Industry Watch
Defense · Command and Control Software
Army awards General Dynamics Mission Systems $99 million for Kestrel airspace software
Announced October 29, 2026 · Company announcement; contract awarded in late September
General Dynamics Mission Systems announced a $99 million firm-fixed-price, indefinite-delivery/indefinite-quantity U.S. Army contract for commercial technology offerings. The work centres on Kestrel, its airspace command-and-control software, which provides services and workflows for the Army's Next Generation Command and Control airspace and mission-planning needs. The company describes Kestrel as hardware-agnostic and containerized, so it can be integrated into larger cloud-based systems or run on hardware a unit already has. The contract itself was awarded in late September; the October 29 announcement is the company's public disclosure of it.
Why it matters: Airspace deconfliction is where crewed aircraft, uncrewed systems, fires, and air defence meet — software that runs on existing unit hardware lowers the fielding barrier for that coordination layer.
Read the source: GDMS awarded $99M U.S. Army contract for Kestrel (ASDNews, from the company announcement)
Defense · Airborne Early Warning
Saab radar flies an early-warning test aboard GA-ASI's uncrewed MQ-9B
Announced October 28, 2026 · Joint company announcement via PR Newswire
Saab and General Atomics Aeronautical Systems said they have flight-tested an unmanned airborne early-warning capability that pairs Saab's LoyalEye radar with the MQ-9B remotely piloted aircraft. The flight, at GA-ASI's Desert Horizon facility in Southern California, was run in a representative operational scenario with stakeholders watching in real time, and the companies said it validated detecting, tracking, and classifying airborne targets from the unmanned platform. The test continues an evaluation programme whose first flight was in May 2026. The companies position the combination as a complement to crewed early-warning aircraft — able to operate alongside them or independently — for persistent surveillance where endurance, reach, and keeping aircrew out of the area matter most.
Why it matters: Early warning has historically required large crewed aircraft; demonstrating the sensing job on a long-endurance uncrewed platform offers a second, more persistent layer rather than a replacement.
Read the source: Saab and GA-ASI successfully flight test LoyalEye AEW capability on MQ-9B (PR Newswire, syndicated)
Daily brief — Wednesday, October 28, 2026
Hardware & AI
AI Accelerators · Inference
Intel names its inference accelerator: GPU D-Series, first card D270
Announced October 12, 2026 · Open Compute Project Global Summit
Intel has given its inference accelerator, formerly code-named Crescent Island, a product name and final specifications. The first card, the GPU D270, is built on Intel's Xe3P architecture and is rated at 1.3 PFLOPS of FP8 compute. Intel's own card carries 160 GB of LPDDR5X memory, and partner-built cards can go up to 480 GB, with 1.5 TB/s of memory bandwidth. The whole design fits in a 350 W, air-cooled PCIe Gen5 x16 card. Customer sampling is underway; Intel has announced neither pricing nor a general-availability date.
The interesting choice is the memory. Instead of the high-bandwidth memory (HBM) used by flagship training accelerators, Intel bet on memory capacity — a great deal of lower-cost LPDDR5X — on the theory that inference economics are driven by how large a model one card can hold and serve, not by peak bandwidth alone.
Why it matters: Inference is where AI hardware spending is heading, and this is a credible capacity-per-watt design point that fits standard air-cooled servers — no liquid-cooling retrofit required to evaluate it.
Read the sources: Intel GPU D-Series specifications (LavX News) · GPU D270 specs and availability explained (Pandromeda)
AI Accelerators · Vendor Specifications
NVIDIA's published Vera Rubin bandwidth figures have been revised down
Checked October 27, 2026 · Vendor specification pages
NVIDIA's current official specifications for the Vera Rubin NVL72 list per-GPU HBM4 memory bandwidth of 19.2 TB/s and GPU-to-GPU NVLink bandwidth of 3 TB/s. Those figures are about 13% and about 17% lower, respectively, than the numbers in an official NVIDIA technical explainer published on July 21. HBM4 capacity per GPU is unchanged at 288 GB. Capacity and bandwidth are different metrics, and the revision is in bandwidth — the rate at which data moves — not in how much memory each GPU carries.
Why it matters: Vendor specification pages are living documents. Any design, bid, or capacity plan built on a bandwidth figure should cite the figure and the date it was read — a number quoted from July is no longer the number on the page in October.
Read the source: NVIDIA Rubin specification changes, NVL72 and HGX (XenoSpectrum)
Memory & Supply Chain
Memory · HBM
HBM4E samples are in customers' hands — before any standard exists
Reported October 2026 · Vendor announcements
Samsung and SK hynix have both shipped 12-high HBM4E samples to major customers. HBM4E keeps the HBM4 interface width and pushes per-pin data rates to roughly 14–16 Gb/s, which works out to about 3.6 TB/s of bandwidth per stack. There is no JEDEC standard for HBM4E yet — as of October 2026 the published specifications cover HBM4, so HBM4E is currently a vendor-defined generation with no independent benchmarks. Volume production is expected in 2027, and it arrives into a tight market: Samsung has said its 2026 HBM4 volume is effectively sold out.
Why it matters: The memory roadmap is being set by vendor silicon ahead of the standards process, while current-generation supply is already allocated. Programs planning AI hardware refreshes should treat HBM supply — not GPU supply — as the long pole.
Read the source: HBM4E: what Samsung and SK hynix's new AI memory does (RCR Tech)
Memory · Supply Outlook
Micron: HBM4 in high-volume production, a new generation every year to 2030
Status as of October 26, 2026 · Vendor roadmap
Micron's HBM4 is in high-volume production: the 36 GB 12-high part is shipping and the 48 GB 16-high part is sampling with multiple customers. Micron's roadmap lays out a new HBM generation every year through 2030 — HBM4E in 2027, then HBM5, HBM5E, and HBM6. The supply picture stays tight: Micron has said its 2026 HBM supply was sold out, and that supply will remain tighter than demand through 2027 and 2028.
Why it matters: A named, dated vendor roadmap plus a sold-out current generation means memory allocation decisions made in 2026 will still be binding on programs delivering in 2028. Design for the memory you can actually get allocated.
Read the source: Micron HBM roadmap: a new generation every year to 2030 (TrendInTech)
Defense & Industry Watch
SALAR Technologies tracks defense and aerospace electronics — the fields its published research and engineering services address — so we watch accelerator platform timelines the way we watch component lifecycles: a platform that slips, or a specification that changes, moves real program schedules. All figures in this section are vendor-stated and are labeled as such.
Industry Watch · Accelerator Platforms
The accelerator field, by memory: AMD MI455X and Microsoft Maia 300
As of October 16, 2026 · All figures vendor-stated
AMD's Instinct MI455X lists 432 GB of HBM4 memory at 23.3 TB/s of bandwidth (vendor-stated). AMD says its Helios racks — 72 MI455X accelerators per rack — are now in production, and first deployments are expected in the fourth quarter of 2026 (vendor-stated). Microsoft announced its Maia 300 accelerator on October 12, 2026, with 288 GB of HBM4 at 18 TB/s (vendor-stated); Microsoft has stated no availability date. For scale, the Intel D270 above takes the opposite trade: Intel's own card carries 160 GB and partner cards can reach 480 GB, in a 350 W envelope — a capacity-first design, rather than the bandwidth-first HBM designs in this item. No power figure for the MI455X or the Maia 300 is given in the sources for this item, so no power comparison is made here.
Why it matters: Hyperscalers are now fielding their own silicon alongside the merchant vendors. For defense and aerospace programs buying AI compute, that means more platform choices — and more lifecycle and support models to evaluate before committing.
Read the source: AI accelerators with the most HBM per chip, October 2026 (Metaprompt)
Additional items — Wednesday, October 28, 2026 (later update)
Added by the daily update after the first edition above was already posted today; the first-edition items are unchanged.
Hardware & AI
AI Infrastructure · Cloud Silicon Choice
Oracle widens its cloud chip menu: Google processors and Cerebras inference join an expanded Nvidia line
Announced October 27, 2026 · Reported by Reuters
Oracle said it will expand its Nvidia-based cloud capacity to multi-gigawatt scale using Vera Rubin systems, while also offering silicon from other suppliers for the first time. Google's Tensor Processing Units will become an Oracle Cloud Infrastructure option, and Oracle also plans offerings from Cerebras Systems and SambaNova Systems, both aimed at inference work. In a further sign of mixing suppliers, Oracle said even its Nvidia-based systems will use Oracle's own Acceleron networking chips in place of some Nvidia networking parts. No value for the expanded Nvidia arrangement was disclosed in the report.
Why it matters: A major cloud is treating accelerator choice as a per-workload decision rather than a single-vendor commitment — which changes the comparison buyers have to make from peak chip performance to fit, throughput, and economics for each workload.
Read the source: Oracle expands chip supply deal with Nvidia, adds Google and Cerebras offerings (Reuters, syndicated)
Edge AI · Neural Processing IP
Ceva rolls one neural-processor family from always-on sensors up to data-center inference
Announced October 28, 2026 · Company announcement
Ceva introduced a next generation of its NeuPro neural processing portfolio, organized as four products — Nano, Prime, Max, and Ultra — built on a common architecture with a single software development kit. The company states the new generation delivers up to twice the AI throughput per square millimetre and 50% better energy efficiency than the first generation (vendor-stated). The line spans always-on sensing at the low end through to Ultra, which targets large-model inference in cloud and data-center settings and integration into memory and networking silicon. Ceva said the architecture is available now to select early-access customers, with general availability planned for the second quarter of 2027.
Why it matters: A single licensable architecture and software stack across sensor, edge, and infrastructure tiers lets chip designers reuse one development path as products scale — a different route to inference capacity than buying finished accelerators.
Read the source: Ceva unveils next-generation NeuPro portfolio (Kalkine, from the company announcement)
Memory & Supply Chain
Memory · HBM Demand
SK hynix says customers are asking for more than twice the HBM it can supply
Stated October 28, 2026 · Third-quarter earnings call
On its third-quarter earnings call on October 28, SK hynix said the volume of high-bandwidth memory customers are requesting exceeds twice what the company can supply, and that its medium- to long-term outlook for HBM demand is unchanged. The company also addressed a move by some customers to cut HBM stacks from twelve layers to eight to ease memory cost pressure. SK hynix said that change would not shrink total demand on a bit basis: the same capacity in lower stacks means more packages, which in turn raises demand for base dies and back-end processing.
Why it matters: Requests running at more than double available supply confirm HBM allocation — not accelerator supply — remains the binding constraint, and thinner stacks shift the bottleneck to packaging rather than removing it.
Read the source: SK hynix says HBM orders exceed twice its supply capacity (Seoul Economic Daily)
Memory · Supply Chain
China's CXMT prepares limited HBM3E output despite reported low yields
Reported October 28, 2026 · Reuters reporting
ChangXin Memory Technologies, China's largest DRAM maker, plans to begin limited production of HBM3E memory within the next two months, according to Reuters reporting published October 28. CXMT has already shipped HBM3E samples to several domestic AI chip designers. Reported yields are 20% to 30% — the first specific yield figure for the part to be made public, per the report — compared with around 70% reported for SK hynix's comparable memory, so early volumes are not expected to translate quickly into meaningful extra supply. The report also notes that Samsung and SK hynix have moved on to mass production of the next HBM generation, HBM4, leaving a gap of more than one generation even after CXMT's HBM3E starts.
Why it matters: Yield, not the production announcement, decides how much usable advanced memory actually reaches the market — at the reported yields, each finished stack carries the cost of several discarded ones.
Read the source: CXMT plans limited HBM3E production within two months, Reuters reports (Silicon Report)
Defense & Industry Watch
Space · Nuclear Propulsion
Kratos to build the power conversion system for NASA's fission-powered SR-1 Freedom spacecraft
Announced October 28, 2026 · Company announcement
Kratos Defense & Security Solutions announced a $33.6 million contract award to design, build, and deliver the power conversion system for NASA's Space Reactor-1 Freedom spacecraft, working as a subcontractor to Sierra Lobo under NASA's SpaceDOC III contract. Its system is a compact, space-qualified closed Brayton cycle unit that converts heat from the onboard fission reactor into electrical power for the spacecraft's propulsion thrusters and other systems. SR-1 Freedom is targeted for launch in 2028 and is intended to travel beyond Earth orbit to Mars, demonstrate nuclear electric propulsion, and deliver the SkyFall payload. The company describes it as the first nuclear fission reactor-powered interplanetary spacecraft.
Why it matters: Power conversion is the link that turns a space reactor into usable propulsion power — awarding that hardware moves fission-electric propulsion from concept work toward a flight program with a launch target.
Read the source: Kratos receives contract for SR-1 Freedom power conversion system (GlobeNewswire)
Defense · Missile Warning Software
Space Force missile-warning ground system takes in a new geostationary sensor feed
Announced October 28, 2026 · Company announcement
SciTec, a Firefly Aerospace company, announced an operational milestone for the U.S. Space Force's FORGE ground system: data from the Space Systems Command's Wide Field of View space vehicle in geostationary orbit has now been incorporated into the system. The sensor provides overhead persistent infrared coverage, and its data products have been integrated alongside data SciTec already processes from other missile warning and tracking satellites. SciTec said the capability is being delivered directly into operational workflows for Space Force Guardians, under its $259 million FORGE Enterprise OPIR Solution contract as stated in the announcement. The company credited FORGE's open architecture with allowing a new sensor stream to be added without infrastructure changes.
Why it matters: Missile warning depends on fusing new sensors into the ground software without rebuilding it — each feed integrated this way widens persistent coverage while keeping the operational system intact.
Read the source: SciTec achieves operational milestone for U.S. Space Force FORGE system (GlobeNewswire)