Morning Digest — November 19, 2026
Satellite-to-phone connectivity is shifting from demonstrations into standards-based services, while frontier AI is beginning to accelerate both its own research process and the hardware beneath it.
Starlink Mobile launches in Mongolia
Starlink and Mobicom have launched mobispace, a direct-to-phone service for Mongolia. In areas without terrestrial cell signal, compatible phones can automatically connect to Starlink satellites to use messaging apps and SMS.
The rollout matters because it brings direct-to-cell service to one of the world’s least densely populated countries, where extending terrestrial towers across remote terrain is costly. It is another concrete step from satellite messaging trials toward routine mobile-network integration.
Source: Starlink
Europe demonstrates regenerative 5G NR-NTN from orbit
i2CAT established what it describes as Europe’s first direct, stable 5G NR-NTN connection through a regenerative satellite. The link operated at 2 GHz between an adapted mobile terminal at its Móra la Nova ground station and the 6GStarLab LEO satellite.
Unlike a transparent relay, 6GStarLab integrates a 3GPP-compliant 5G base station into the satellite. Processing and routing traffic onboard can lower latency, increase capacity, reduce dependence on ground stations, and provide an operational testbed for architectures relevant to Europe’s IRIS² constellation.
Source: i2CAT
ESA and SES launch a standards-based direct-to-device project
ESA and SES signed the SELYCE Direct-to-Device Partnership Project to develop and test an end-to-end satellite connectivity system for everyday devices. Initial applications include connected vehicles, messaging and emergency voice beyond mobile coverage, and resilient communications during disasters.
The project sits explicitly within converged terrestrial and non-terrestrial 5G and future 6G architectures. Industrial work is led by SES with European research and technology partners, while Luxembourg, Belgium, and the United Kingdom provide primary public funding.
Source: European Space Agency
Anthropic prices its IPO at $83 a share
Anthropic priced its initial public offering at $83 per Class A share after the registration statement became effective on November 18. The company expects trading to begin on Nasdaq today.
The listing creates a rare public-market benchmark for a frontier-model developer. Investors will now be able to judge the economics of model development, enterprise adoption, and compute spending with more transparency than private funding rounds provide.
Source: Anthropic
OpenAI proposes tracking recursive research acceleration
OpenAI published new internal data showing that researchers’ use of AI continues to grow and that agents are completing longer tasks, although humans remain deeply involved. Human work on safety and alignment is also growing faster than effort in any other individual research area.
The governance proposal is as important as the productivity finding: OpenAI argues that frontier labs should track and publicly report indicators of recursive self-improvement—AI accelerating the development of future AI systems—so the industry can detect and govern the transition rather than infer it after the fact.
Source: OpenAI
OpenAI models improve AMD GPU designs by up to 14%
OpenAI says its models found power, performance, and area improvements of up to 14% on already heavily optimized AMD flagship GPU IP in less than five months. The models rewrote hardware-description code and surfaced improvements beyond years of conventional engineering optimization.
The same approach improved OpenAI’s Jalapeño inference chip, with gains verified on real silicon. OpenAI is also working with Synopsys to connect model capabilities to established electronic-design-automation and validation tools, potentially making AI-assisted chip design accessible to more engineering teams.
Source: OpenAI
NVIDIA ships RTX Spark compact AI desktops
NVIDIA announced availability of RTX Spark compact desktops from ASUS, Dell, and MSI. The dedicated Windows systems support local AI agents, image and video generation, and model fine-tuning with up to 128 GB of unified memory.
Spark targets the space between conventional creator PCs and data-center systems: enough memory for substantial local models, packaged as an OEM desktop rather than specialist infrastructure. That could broaden private, low-latency AI development without requiring every workload to run in the cloud.
Source: NVIDIA
Research Radar
Continuum: Boundary-State-Aware Deterministic Scheduling in Dynamic LEO Networks
Siqi Yang, Zonghui Li, Chaoqun You, and Yue Gao propose carrying buffered-node and arrival-time state across LEO topology transitions instead of discarding unfinished forwarding progress at every snapshot. Their rolling-horizon scheduler improved schedulability by up to 16.6 percentage points and reduced mean jitter by 10.9–37.7 times, with workloads scaling to 10,000 flows.
Source: arXiv 2611.24021
Slice Isolation-as-a-Service: An Isolation Mechanism to Proactively Defend Cross-Slice Attacks in 5G-Advanced and Beyond
Abdulaziz Abdulghaffar and Ashraf Matrawy model strict physical and virtual slice isolation as an on-demand service. In a Free5GC and Nokia Containerlab testbed, shared and VNF-only isolation suffered up to 30% packet loss and full bandwidth starvation under flooding and GTP-in-GTP reflection attacks, while MINLP-driven physical isolation preserved 99.999% reliability.
Source: arXiv 2611.23826
Conditional What-If Generation for Radio Access Network Energy under Rare and Unseen Contexts
Kevin Wardakhan, Marie Line Alberi-Morel, Jérémie Leguay, Afef Feki, Adrien Sardi, Sébastien Archeveque, and Claudiu Mihailescu develop conditional generators for evaluating 5G RAN energy policies when measurements are sparse. Their prediction-based method generalizes to rare and unseen configurations and, with one target observation, reduces median error by about 50% compared with TimeWeaver.
Source: arXiv 2611.24208
Source Notes
IEEE Xplore and ACM Digital Library returned no newly indexed eligible wireless papers. The research section therefore uses verified arXiv records submitted November 18.
Takeaway
Satellite-to-phone connectivity is moving from isolated trials into deployable, standards-based systems just as AI begins closing the loop on its own research and hardware design.