Castles or Solid Foundations?

Castles or Solid Foundations?

Hello and welcome,

Can infrastructure keep up with AI's appetite, or are we building castles on shaky foundations? NVIDIA and OpenAI push inference economics to the center of commercial viability, while wireless communication chipsets are expected to nearly double to $37.88 billion by 2031. Meanwhile, OQ Technology and Eseye merge satellite and terrestrial networks into single connectivity ecosystems, and 5G standalone architecture reaches 85 operators across 47 markets. The technology sector is experiencing a fundamental shift where compute capacity, energy availability, and connectivity infrastructure determine which AI applications scale and which remain trapped in pilot purgatory.

"The convergence of inference economics, 5G standalone maturity, and hybrid terrestrial-satellite connectivity shows an industry transitioning from proof-of-concept demonstrations to addressing the practical constraints that determine commercial success at scale."

Article content

Inference Economics Becomes AI's New Battleground

NVIDIA invested in Baseten, an AI startup focused on running models efficiently in production, signaling that the era of training mega-models is giving way to the era of serving them at scale. The investment underscores rising value in companies addressing "inference" how AI systems respond to user queries after training completes. Latency, cost per query, and reliability now decide whether AI features actually make economic sense for enterprises rather than remaining expensive experiments.

Analysts increasingly expect inference workloads to dominate overall AI compute as more products integrate assistants, copilots, and automation across customer support, sales, analytics, and internal operations. Training massive foundation models captures headlines, but the real commercial challenge is answering millions of queries daily at acceptable cost and speed. Baseten's pitch addresses production bottlenecks that prevent AI systems from scaling beyond demos, issues like model optimization, deployment complexity, and infrastructure management that eat engineering time and operational budgets.

What makes this moment significant is the shift from capability demonstrations to economic viability. OpenAI disclosed a renewable energy agreement to secure long-term power for data center expansion, highlighting that compute capacity is now gated by energy availability and grid readiness rather than just GPU access. Large model providers are increasingly acting like industrial-scale power buyers, negotiating multi-year arrangements that resemble hyperscalers' playbook. When AI leaders can guarantee power and uptime, they can sign bigger enterprise contracts, run more inference, and train larger next-generation systems. The infrastructure layer is becoming a competitive moat that separates companies with serious deployment plans from those still pitching vaporware.


Wireless Communication Chipsets Market Doubles on 5G and IoT Demand

The global wireless communication chipset market is projected to expand from $19.17 billion in 2025 to $37.88 billion by 2031, representing 12.02% compound annual growth. These specialized integrated circuits transmit and receive data across cellular networks, Wi-Fi, and Bluetooth, serving as fundamental hardware for connectivity standards. The upward trajectory is driven by aggressive 5G infrastructure rollout, extensive IoT incorporation across industrial sectors, and rising demand for high-bandwidth mobile computing.

5G standalone architecture deployment acts as the primary growth engine. As telecommunications operators shift toward 5G SA networks and adopt millimeter-wave spectrum, demand surges for advanced baseband processors and RF front-end modules capable of handling complex frequency combinations. Global 5G subscriptions reached approximately 2.4 billion in Q1 2025, while worldwide semiconductor sales hit $627.6 billion in 2024, reflecting massive hardware demand generated by communication technologies. The infrastructure expansion directly links to subscriber adoption, necessitating high-volume chipset production for both user equipment and base stations.

The practical impact shows up across multiple layers of the technology stack. Wi-Fi 7 chipsets enable speeds exceeding 40 Gbps with multi-link operation for simultaneous transmission across frequency bands. 5G RedCap modules create a mid-tier connectivity option between low-power LPWAN and high-performance 5G NR, with enhanced RedCap launches expected throughout 2026. The automotive segment represents particularly robust growth, with connected vehicle subscriptions forecast to rise from 500 million today to 1.2 billion by 2035. Software-defined vehicles requiring cellular connectivity for over-the-air updates, telematics, and safety features are driving this expansion. The chipset market isn't just growing, it's restructuring around standards that prove themselves in real deployments at scale.


Satellite and Terrestrial Networks Merge Into Single Connectivity Fabric

OQ Technology entered a strategic partnership with UK-based IoT connectivity specialist Eseye to deliver seamless, worldwide 5G IoT connectivity. The agreement combines satellite and terrestrial networks to provide reliable, always-on connectivity for IoT devices, including in remote and hard-to-reach locations. The collaboration integrates OQ Technology's Low Earth Orbit satellite constellation with Eseye's AnyNet Connectivity Hub, creating a single connectivity ecosystem supported by one SIM solution.

Using the 3GPP Release 17 standard, the partnership enables IoT devices to seamlessly switch between cellular and satellite networks. This represents more than incremental improvement, it addresses a fundamental constraint where terrestrial cellular networks cover only a fraction of Earth's surface, leaving vast regions unreachable for connected devices. Industries like maritime shipping, agriculture, mining, and logistics operate equipment that moves beyond cellular coverage regularly. These sectors previously chose between expensive satellite-only solutions or accepting connectivity gaps. The hybrid approach eliminates that trade-off.

The timing matters because 5G non-terrestrial networks are transitioning from experimental trials to commercial reality. Frankfurt Airport deployed Europe's largest private 5G campus network covering more than 20 square kilometers, exemplifying the scale at which organizations invest in dedicated wireless infrastructure. At least 85 operators across 47 markets launched commercial 5G standalone networks by end of 2025, with network slicing commercialization enabling distinct performance characteristics over single physical infrastructure. The convergence of terrestrial 5G, satellite 5G NTN, and edge computing creates truly global IoT coverage where devices roam between networks based on location and requirements. Companies no longer choose between terrestrial and satellite connectivity, they deploy solutions using both, switching automatically based on availability, latency requirements, and cost optimization.


Article content

January 27, 1880

January 27, 1880 Thomas Edison received U.S. Patent No. 223,898 for his electric lamp, embodying the principles of his incandescent design that paved the way for universal domestic use of electric light.

Edison didn't invent the first electric light, incandescent lamps with carbon illuminants existed 30 years prior. What Edison achieved was creating a thin carbon filament of high resistance within a practical electrical illumination system that offered critical advantages over earlier designs using thick carbon rods. The filament's draw of small currents enabled sufficiently thin copper supply wires to be deployed at low cost. More importantly, it enabled networks of many lamps to be electrically connected in parallel rather than series, making each lamp's continuous operation independent of others when they burned out.

Edison and his team at Menlo Park tested over 6,000 materials before discovering that carbonized bamboo produced a filament lasting more than 1,200 hours, compared to just a few hours for earlier attempts. The breakthrough came after Edison spent time absent-mindedly rolling compressed carbon between his fingers, leading him to carbonize materials systematically. He tested every plant imaginable including baywood, boxwood, hickory, cedar, flax, and bamboo. The persistence paid off when he achieved a reliable, long-lasting light source that could operate 50 to 60 days continuously.

Edison's patent represented more than protecting a single invention, it laid the foundation for an entire electrical distribution system. He founded the Edison Illuminating Company on December 17, 1880, and patented systems for electricity distribution throughout the decade. His approach transformed electric lighting from experimental curiosity to commercially viable technology that replaced gas lamps. The patent also sparked fierce competition and legal battles, including disputes with Joseph Swan in Britain that led to forming a joint company called Ediswan. Edison's work with light bulbs propelled the United States out of the gaslight era and into the electric age, fundamentally changing how humanity experiences night and enabling the 24-hour industrial society that followed.


Article content

Did you know?

The 5G Standalone Inflection Point

At least 85 operators across 47 markets launched commercial 5G standalone networks by the end of 2025, marking a transition from deployment phase to operational maturity. This acceleration is underpinned by growing industry confidence that 5G SA is now sufficiently mature to support network slicing at scale and guarantee specific service level agreements. The distinction matters because early 5G deployments ran on 4G infrastructure, limiting the technology's transformative capabilities.

5G standalone architecture enables the features that separate 5G from simply being faster 4G. Ultra-reliable low-latency communication achieves response times under 1 millisecond, necessary for autonomous vehicles and robotics where tiny delays could cause accidents. Network slicing delivers multiple virtual networks with distinct performance characteristics over single physical infrastructure, addressing the fundamental challenge of serving vastly different use cases from ultra-reliable low-latency communications to massive IoT deployments without building separate networks for each. Enhanced mobile broadband delivers peak data rates exceeding 10 Gbps, supporting bandwidth-intensive applications like 4K video streaming and AR/VR experiences.

The maturation of network slicing has significant implications for spectrum efficiency and capital expenditure, potentially allowing operators to extract greater value from existing infrastructure investments whilst meeting diverse enterprise requirements. Capgemini's assertion that programmable network software and APIs will become a necessity from 2026 onwards aligns with joint ventures like "Aduna" formed by Ericsson and global telecommunications providers. Dell'Oro Group's analysis indicates that the Radio Access Network market remained largely stable in Q3 2025, with growth in Europe, the Middle East, and Africa offsetting declines in North America and Asia Pacific. The infrastructure is stabilizing as focus shifts from building networks to monetizing them through enterprise services, private 5G deployments, and advanced applications that were impossible under non-standalone architectures.


Article content

The convergence of terrestrial and satellite IoT connectivity represents more than incremental improvement, it's restructuring how companies think about global deployments. Traditional approaches forced binary choices between cellular networks offering high bandwidth in limited geography versus satellite systems providing global coverage at lower data rates and higher costs. The hybrid model emerging in 2026 eliminates that trade-off through seamless switching protocols defined in 3GPP standards.

Technical implementation relies on integrated SIM solutions that authenticate across both network types, dynamic routing algorithms that select optimal connectivity based on device location and application requirements, and edge computing layers that preprocess data locally before transmission to minimize bandwidth consumption. OQ Technology's LEO satellite constellation operates in L-band spectrum, providing coverage where terrestrial networks don't reach while maintaining compatibility with standard cellular IoT protocols. Eseye's AnyNet Connectivity Hub manages this complexity, presenting a unified interface to applications regardless of underlying transport.

The business implications extend across multiple industries. Maritime shipping companies can monitor container conditions continuously rather than only when vessels are near shore. Agricultural operations track equipment and environmental sensors across vast properties where cellular coverage is spotty or nonexistent. Mining companies maintain connectivity with autonomous vehicles and safety systems in remote locations. Energy companies monitor pipelines and remote infrastructure without deploying dedicated communication systems. The hybrid approach delivers the coverage of satellite with the economics and performance of cellular where available, fundamentally changing the calculation for deploying IoT at global scale. Companies that previously couldn't justify connectivity costs for certain use cases now find the hybrid economics work, expanding addressable markets for IoT solution providers while creating new operational capabilities for end users.


Article content


That's all for now folks


We examined how inference economics is reshaping AI's commercial viability, wireless chipset markets doubling on 5G and IoT demand, and satellite networks merging with terrestrial infrastructure into unified connectivity fabrics. Three infrastructure stories about technology moving from capability demonstrations to addressing the practical constraints that determine success at scale.

Till next time,

stay connected,

Iliana & the Apiro Data team.



Article content


Insightful take! Scaling AI isn’t just about models, it’s about infrastructure, energy, and connectivity. Excited to see how the industry tackles these bottlenecks while keeping sustainability in mind.

The result is unfortunately a big "No" Apiro Data. There is a global shortage of Memory as supplies of it are being redirected to the highest payers which are the AI players. As a result prices of laptops etc. for us mere mortals are skyrocketing. Also our wonderful push for net zero driven by power generation from so-called renewables means there isn't enough stable power to drive AI. Renewable energy entering power grids is not stable and can take whole countries down as Spain saw last year. As a result us mere mortals again will be impacted as ultimately our personal energy usage will have to be capped to allow AI to grow as all seem to desire. Sorry I am not positive about any of this! 😉

To view or add a comment, sign in

Others also viewed

Explore content categories