Morning Digest — October 28, 2026

Today’s signal is infrastructure: satellite spectrum is consolidating, direct-to-device competition is widening, hyperscalers are scaling AI networks into the gigawatt era, and frontier labs are testing both public markets and new ways to validate AI-generated science.

FCC clears Amazon Leo’s $11.5 billion Globalstar acquisition

The FCC approved Amazon Leo’s acquisition of Globalstar, giving Amazon control of roughly 25 MHz of mobile-satellite spectrum plus related network assets. The move advances Amazon’s direct-to-device ambitions alongside its broadband constellation and turns the satellite-phone market into a sharper contest among Amazon, Starlink, AST SpaceMobile, and other emerging operators.

Source: Light Reading

AST SpaceMobile expands its direct-to-device footprint in Brazil

Brazilian regulator Anatel approved AST SpaceMobile’s S-band spectrum use and secondary access to the 700 MHz band. AST also announced agreements with four regional mobile operators and plans for local ground infrastructure, broadening the path toward standards-based satellite broadband for ordinary phones in one of the world’s largest mobile markets.

Source: AST SpaceMobile

OpenAI releases three AI-generated claims for expert scrutiny

OpenAI says its internal systems generated major results in three fields: a proof of local smoothing with implications for Kakeya and restriction conjectures, a refutation of the Exponential Time Hypothesis, and existence results for interacting quantum field theories including pure Yang–Mills. Because claims of this magnitude require far more than machine-produced formal artifacts, OpenAI is handing materials to independent academic groups and experimenting with different release and verification models.

Source: OpenAI

Anthropic files for an IPO under the ticker ANT

Anthropic filed a Form S-1 for a proposed Nasdaq Global Select Market listing under “ANT.” The company has not yet set the offering size or price range; Goldman Sachs and Morgan Stanley are named in the prospectus process. If completed, the listing would give public markets unusually direct exposure to a frontier-model developer while increasing pressure for financial and governance transparency.

Source: Anthropic

NVIDIA and Oracle plan multi-gigawatt Vera Rubin superclusters

Oracle Cloud Infrastructure is making NVIDIA Vera Rubin NVL72 systems and Vera CPUs generally available and plans to expand its superclusters to multi-gigawatt scale. The design combines rack-scale GPU systems, Spectrum-X Ethernet, BlueField DPUs, and Oracle’s Acceleron network while letting customers scale CPU and GPU capacity independently—important for agent workloads whose tool execution and data retrieval can leave expensive GPUs idle.

Source: NVIDIA Newsroom

Meta opens more of its gigawatt-scale AI networking stack

Meta’s OCP disclosures span the network from rack to region: 102.4-terabit liquid-cooled switch racks, a flattened scale-out fabric with eight network planes per GPU, petabit-scale “scale-across” connectivity, and Ethernet scale-up systems for in-house accelerators. Meta also says MetaRoCE will become open source and is co-leading an OCP effort with Microsoft and Alibaba to standardize the switch platform layer.

Source: Meta Engineering

U.S. Treasury proposes broader outbound-investment limits on China

The proposed expansion would tighten outbound-investment controls covering semiconductors and microelectronics, artificial intelligence, quantum technologies, hypersonic systems, and selected high-performance-computing inputs. It would also broaden restrictions involving designated foreign entities, potentially adding new compliance burdens for U.S. investors and technology companies operating across complex ownership structures.

Source: WilmerHale

NVIDIA launches Minerva to operate AI factories

At GTC Berlin, NVIDIA introduced Minerva, an AIOps platform intended to surface infrastructure problems earlier and keep AI factories productive. The launch signals a maturing bottleneck: after organizations acquire accelerators and power, reliable operation, fault localization, utilization, and fleet-level observability determine whether that capacity produces useful work.

Source: NVIDIA Blog

Research Radar

Artur Magamedov, Artem Krasilov, Roman Zlobin, and Evgeny Khorov propose a multi-level outer-loop link-adaptation method that assigns different reliability targets across retransmission attempts. In ns-3 simulations, it raised system capacity by up to four times and reduced the minimum latency budget for 99.9% reliability from 60 ms to 40 ms. The paper is accepted at IEEE GLOBECOM 2026.

Source: arXiv

Agentic Digital Twin-Assisted Multimodal Predictive Handover

Afan Ali, Ali Arshad Nasir, Naveed Iqbal, and Daniel Benevides da Costa combine a synchronized multimodal digital twin with agents for forecasting, handover planning, and feedback adaptation. Their approach reduced average handovers from 11 to 8 per trajectory while maintaining 98.2% ray-path availability in dense mid-band and mmWave scenarios.

Source: arXiv

Multi-view Semantic-aware Resource Allocation for Edge AI

Jiawei Luo, Hussein Saleh, and Mohamad Assaad formulate joint semantic compression, transmission-mode selection, and physical-resource-block assignment for multi-user edge AI. Their Lagrangian-relaxation method approached optimal simulated utility while doubling PRB efficiency compared with average allocation; the manuscript is submitted to IEEE Wireless Communications Letters.

Source: arXiv

Source Notes

IEEE Xplore and ACM Digital Library searches returned no newly indexed matching papers. The research section therefore uses verified arXiv records, including one accepted IEEE conference paper and one IEEE journal submission.

Takeaway

Competition is shifting from isolated AI models and satellite launches toward the infrastructure, spectrum, networks, and operating systems that make intelligence and connectivity usable at scale.