The Pulse™ | Week 26: Capital picks the layer
Five signals from the fog. Coverage window: June 15 to 19, 2026.
By Armando Pereira | Founder, PVentures Consulting | Senior Member IEEE | Co-founder, OpenFog Consortium (IEEE 1934) | President, Autonomous Vehicle Computing Consortium | Former VP/GM Optical BU, Centillium Communications (CMOS PON SoC, NTT-qualified)
👋 Welcome back to The Pulse™
The Pulse™ is your weekly operating brief for execs, founders, and investors. What changed across Industrial IoT, Telecommunications, Edge Computing, Autonomous Systems, and Artificial Intelligence, why it matters, and what to do next, all in one place.
🩺 This Week’s Pulse™
This week, the market stopped paying for the application and started paying for the layer underneath it.
A neocloud entered the Nasdaq-100, forcing more than $ 800 billion in index funds to buy it.
Turbine slots for AI data centers are booked into 2030.
Spectrum, radios, and the factory floor drew the same capital.
Across all five verticals, scarcity moved to whoever owns the physical and financed substrate; everyone else has to rent. The question for leadership is no longer which model wins. It is which layer you own.
Share this: In 2026, AI value is migrating from the model to the layer beneath it: power, spectrum, silicon, and the financed cloud. Own the layer, or rent it.
🏭 INDUSTRIAL IOT: The factory floor becomes a private network
Siemens detailed an expansion of its private 5G infrastructure to the United States and seven more countries, pairing CBRS-band radios with 5G routers that run applications on an onboard edge runtime. It is the clearest sign yet that the industrial network, not just the industrial robot, is the asset.
Siemens private 5G reaches US manufacturers on the CBRS spectrum, with the edge runtime on the router itself.
The Siemens and NVIDIA Industrial AI Operating System names Erlangen, Germany, as the first fully adaptive factory blueprint.
Industrial power is now a capital contest: GE Vernova booked $2.4 billion in data-center electrification orders in a single quarter, with turbine slots full through 2030.
I ran an industrial IoT company in Northern Italy from 2014 to 2017, and the lesson, held then as now, is that the plant that owns its connectivity and its power-schedule controls its own automation roadmap.
📡 TELECOMMUNICATIONS: AI-RAN sets the agenda before MWC Shanghai
With MWC Shanghai opening on June 24 at the Shanghai New International Expo Center, the pre-show agenda already centers on AI-RAN, a dedicated 6G Zone, and satellite non-terrestrial networks. The radio is being rebuilt as an AI workload.
The GSMA program centers on Mobile AI, 5G-Advanced, satellite and NTN, and 6G, per Omdia’s preview.
AI-RAN demand has tripled year over year, with most AI-RAN Alliance demonstrations running on a single accelerated-compute platform.
Carriers are choosing where AI compute sits in the network, and that decision sets the cost structure for a decade.
I worked on the go-to-market for FTTH broadband silicon after the Lantiq spin-out from Infineon in Germany, and carried products through NTT Labs qualification at Marvell, with years on the TIA Board; the carrier value chain rewards whoever owns the qualified layer, not the loudest demo.
⚡ EDGE COMPUTING: Inference splits in two
As capital pours into centralized GPU clouds, a countercurrent is hardening at the machine edge. Qualcomm’s Dragonwing industrial processors now run real inference on-device, and Siemens is shipping edge runtime directly into the network gear.
Qualcomm Dragonwing Q-7790 delivers 24 TOPS of on-device NPU performance for industrial vision and sensors, with no cloud required.
Edge AI and private 5G are converging into a single on-premises stack for real-time factory decisions.
The architecture question is placement: what runs at the edge, what runs in the cloud, and who certifies the boundary.
The fog and edge reference architecture published as IEEE 1934, which I helped author while at OpenFog, specified exactly this split: a validated continuum from device to cloud, not a single location.
🚗 AUTONOMOUS SYSTEMS: The production gap becomes undeniable
The autonomy story this month is no longer about demonstrations; it is about unit economics. Aurora is scaling driverless freight while Tesla’s Texas robotaxi fleet remains a fraction of Waymo’s.
Aurora has logged over 250,000 driverless miles across 10 commercial routes, including a roughly 1,000-mile Fort Worth-to-Phoenix lane that exceeds human hours-of-service limits.
Aurora targets 200 driverless trucks and an $80 million revenue run rate by year-end, at 20 trucks per week in the second half.
Tesla’s Texas robotaxi fleet was reported to be roughly 42 vehicles, compared with Waymo’s nearly 500,000 paid weekly trips across 11 cities.
As AVCC president from 2017 to 2025, I watched OEMs and compute suppliers spend years building the governance and safety table; the companies converting that work into route miles and revenue are the ones that win the decade.
🤖 ARTIFICIAL INTELLIGENCE: $800 billion forced to buy the cloud
The sharpest signal of the week came from the index, not the lab. Nebius, a neocloud that rents GPU capacity, joins the Nasdaq-100 on June 22, forcing passive funds that track more than $800 billion to buy it whether they want to or not.
Nebius touched a record $298.85 on June 18, up roughly 21% since the June 11 inclusion announcement.
The company carries a Meta agreement worth up to $27 billion over five years and a $25 billion capex plan; Microsoft has committed $19.4 billion.
Index mechanics now amplify capital concentration in financed AI infrastructure, independent of any model breakthrough.
The money is pricing the landlord of compute, not the tenant. That is a structural shift, not a trade.
🔥 3 Non-Obvious Takeaways
1. The scarce input moved from chips to schedules
GE Vernova’s turbine slots fill into 2030, and Nebius’s forced index buying is the same story: the binding constraint on AI is no longer silicon supply; it is the power, land, and capital calendar. Boards still optimizing for GPU allocation are fighting last year’s war.
2. The edge is the hedge against the cloud cartel
As index flows and hyperscaler deals concentrate compute in a handful of financed neoclouds, Qualcomm’s on-device inference and Siemens’s edge runtime give industrial operators a way to keep latency-critical and proprietary workloads off the rented layer. Sovereignty at the edge is becoming a procurement strategy, not a slogan.
3. Regulation is quietly resetting the compliance clock
The EU’s June Code of Practice on AI content marking and the Digital Omnibus deadline relief, which pushes high-risk obligations to December 2027, lower near-term compliance costs for the largest model owners while the capital layer consolidates underneath them. The governance clock and the capital clock are now out of sync.
🗺️ The arc so far
This issue is the latest in a running cross-stack thesis on how the AI infrastructure stack is being governed, owned, and now financed. The full arc, in order:
The Sextant™ (May 6): “The Hidden Bottleneck in AI Compute.” Named compute capacity as the binding constraint on the AI buildout.
The Pulse™ Week 20 (May 12): “Physical AI Goes Operational.” Showed the constraint moving onto the factory and city floor.
The Vector™ (May 14): “The Inference ASIC Fork.” Traced the silicon split between training and inference economics.
The Sextant™ (May 20): “The Hidden Lock-In Beneath Inference and Physical AI.” Showed how that split hardens into vendor lock-in.
The Pulse™ Week 22 (May 26): “Capacity Gets an Address.” Located capacity in specific power, land, and grid positions.
The Pulse™ Week 24 (June 9): “The AI Stack Goes Institutional.” Marked the shift from capacity to governance and standardization.
The Pulse™ Week 25 (June 16): “Owning the layer.” Value shifted to whoever owns the defensible layer.
This issue, Week 26 (June 23): Capital markets begin repricing the stack by layer, and money flows to whoever owns the scarce physical and financed substrate.
🧭 Where to start
If which layer of the AI stack your company owns versus rents is now a live board decision, the Board Advisory Session is the right place to pressure-test it before the capital and power calendars close further:
❓ Question for you
Which of your four operational verticals, Industrial IoT, Telecommunications, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting the moment the layer you depend on stops being a commodity and becomes a landlord?
If you are not tracking your power and grid-interconnect lead times, your spectrum and connectivity ownership, and your make-versus-rent ratio on inference compute, you are making 2026 procurement calls without the leading indicators that separate early movers from laggards over the next two quarters.