The Pulse™ | Week 21: The Inference Fork
Five signals from the fog. Coverage window: Monday May 11 to May 15, 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 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™
Inference is splitting into two reinforcing stacks, and the middle is hollowing out. Anthropic’s $200 billion, five-year, 5 GW Google Cloud commitment now sits alongside OpenAI inside more than half of the $2 trillion in cloud backlogs. The same week, five European carriers federated their telco-edge into a sovereign AI fabric, EnCharge shipped an analog in-memory edge accelerator, and Waymo extended its driverless footprint past 1,400 square miles. Hyperscaler concentration and sovereign distribution are both winning. Single-vendor cloud-only AI looks like the stranded architecture in between.
Share this: Hyperscaler AI capex and sovereign edge are both winning this quarter; the architecture in between is the one that gets stranded.
🏭 INDUSTRIAL IOT: The construction boom moved to data centers
IoT Analytics’ May Industrial Macro Pulse confirmed what the construction permits already showed: factory reshoring did not match the narrative, but data center construction did. Annualized US industrial-construction spend on data centers grew from roughly $9.5 billion in January 2020 to $47 billion in January 2026, a fivefold jump on the same five-year clock.
IoT Analytics: data centers are now the dominant industrial buildout, larger than factory reshoring
Schneider Electric, via Microsoft Industrial Copilot on Azure AI, reports up to 50% engineering-time cuts for early adopters
Hannover Messe 2026: Siemens introduced the Eigen Engineering Agent, moving copilot AI from assistance to autonomous execution
Running Wi-Next in Northern Italy a decade ago, the operating question was how to instrument what already existed; in 2026, the new instrument is the building going up next door, and it is full of GPUs, not motors.
📡 TELECOMMUNICATIONS: Five European carriers federate the edge
On May 13 to 15, Vodafone, Deutsche Telekom, Orange, Telefónica, and TIM advanced a federated European telco-edge cloud aimed at sovereign IoT and AI workloads. The same week, T-Mobile and Ericsson published trial results for an AI-native scheduler on a live 5G Advanced network: spectral efficiency up nearly 10%, download speeds up 15%. Nokia handed its mobile unit to a Siemens executive with an explicit mandate to be AI-native-by-design.
Federated European edge: transport logistics, industrial automation, emergency services on EU-based network slices
T-Mobile / Ericsson: commercial deployment of the AI-native scheduler planned for Q3 2026
Nokia: AI-native 5G Advanced and 6G framed as the new operating discipline
The carrier playbook through the Lantiq spin-out and the Marvell NTT Labs qualification cycle was access-and-backhaul; the 2026 playbook is sovereign-compute-and-RAN-AI, and carriers who do not own a federated edge position by year-end will rent it from those who do.
⚡ EDGE COMPUTING: Analog in-memory inference reaches product
EnCharge AI moved its EN100 analog in-memory accelerator from research to shipping form factors for laptops, workstations, and edge devices. The architectural significance is larger than the product: analog in-memory compute targets the watts-per-inference ceiling that has limited local AI on power-constrained hardware. AMD outlined a deeper commitment to embedded microprocessors for industrial edge environments the same week.
EnCharge EN100: analog in-memory inference, two form factors, runs locally without cloud dependency
AMD: embedded edge processor roadmap framed as core, not peripheral, to the AI strategy
The federated EU telco-edge gives EN100-class accelerators a sovereign home that is not a hyperscaler
The fog and edge reference architecture published as IEEE 1934 specified exactly this convergence point: compute and connectivity validated as one stack at the machine edge. Analog in-memory accelerators are the power envelope that the 2017 specification could not yet name.
🚗 AUTONOMOUS SYSTEMS: Waymo expands and recalls in the same week
On May 13, Waymo enlarged its driverless service footprint to over 1,400 square miles across 11 US cities, an area larger than Rhode Island, with Miami first and Austin, Atlanta, Houston, and the San Francisco Bay Area widening behind. The company runs roughly 400,000 paid rides per week against a stated one-million-per-week year-end target. The same week, it voluntarily recalled 3,800 vehicles due to a software issue that could allow cars to drive into standing water.
Coverage: 1,400+ square miles, 11 cities, six are FIFA World Cup hosts (May 13)
Recall: 3,800 robotaxis, software-only fix, no injuries reported (May 12)
Uber publicly committed over $10 billion to Rivian, Lucid, and Nuro alternatives while still carrying Waymo in Austin and Atlanta
Consortia like AVCC spent eight years building the governance table that put OEMs, tier-1 suppliers, and compute-stack vendors in the same room. Waymo’s recall-and-expansion in the same week is the operating reality that the table was built to hold: scaled service and safety transparency cannot run on separate clocks.
🤖 ARTIFICIAL INTELLIGENCE: Two firms now hold half the cloud backlog
The week confirmed Anthropic’s $200 billion, five-year, 5 GW Google Cloud commitment and the accounting reality around it: Anthropic and OpenAI contracts now account for more than half of the roughly $2 trillion revenue backlog disclosed by the major cloud providers. On May 11, OpenAI began rolling out its GPT-5.5-Cyber variant to vetted EU institutions, including the EU AI Office. On May 12, Vodafone’s earnings call candidly stated the constraint: AI is of little use without performant fixed-fiber and mobile 5G networks (networks that deliver high throughput, low latency, low jitter, and low packet loss under real load, with headroom).
Anthropic / Google: 5 GW TPU-based capacity, starting to come online in 2027
OpenAI GPT-5.5-Cyber: EU rollout begins; Anthropic Mythos access still outstanding for EU institutions
Backlog concentration: AI’s two largest buyers now exceed 50% of the $2T hyperscaler revenue book
AI deployment economics is the application end of a market I have been close to since the AVCC AI work; read the commercial structure above as a forward observation, not deep architectural commentary.
🔥 3 Non-Obvious Takeaways
1. The middle of the inference stack is the most stranded position in 2026
Analog in-memory accelerators expand the edge; Anthropic and OpenAI lock in the hyperscaler core. Single-vendor cloud-only inference, which was deployed in 2024 as the enterprise default, is the most exposed configuration heading into 2027. Renegotiate before the dependency hardens.
2. Sovereign compute is now a telecom product, not a policy paper
Five European carriers federated their edge clouds for sovereign IoT and AI workloads the same week Vodafone framed AI as useless without performant fixed and mobile networks. The European carrier industry is positioning itself as the buyer’s alternative to US hyperscaler concentration, with regulated-network SLAs as the differentiator. The next 30 days of contract language will decide whether enterprises take the offer.
3. The Waymo dual signal is the field-deployable AI template
Expanding to 1,400 square miles and recalling 3,800 vehicles in the same week is not a contradiction; it is the operating cadence safety-certified autonomy requires. Any board evaluating an AI deployment that touches physical infrastructure should read Waymo’s week as the working model: scale and recall in the same cycle, transparently, without losing the service rhythm.
🧭 Where to start
If splitting your inference strategy between a hyperscaler core and a sovereign or carrier-edge layer is a live decision for your leadership team in 2026, the Strategic Review Session is the right place to test the question before procurement locks.
❓ Question for you
Which of your four operational verticals, IIoT, Telecom, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting the inference-cost gap between hyperscaler cloud and sovereign or on-prem edge?
If you are not tracking per-inference cost by deployment location, contracted compute-capacity utilization, and SLA latency variance across hyperscaler and edge endpoints, you are making 2026 procurement calls without the leading indicators that will separate the early movers from the laggards over the next two quarters.
🗺️ The arc so far
The Sextant™ (May 6): "The Hidden Bottleneck in AI Compute". Board-level frame; first named the capacity constraint that made the bifurcation visible. The strategic-question anchor for Weeks 19 through 21.
The Pulse™ (May 5, Week 19): "AI Capex Sets the Clock". Quantified the hyperscaler commitment surface that now anchors the Anthropic / OpenAI half of the fork.
The Vector™ (May 7): "The AI-RAN Architectural Fork". Deep dive on the carrier-grade compute choice that the federated EU telco-edge now operationalizes.
The Pulse™ (May 12, Week 20): "Physical AI goes operational". Deployment-wave frame; established that the stack had crossed into production, teeing up the architectural question answered this week.
The Vector™ (May 14): "The Inference ASIC Fork". Silicon-level preview of the architectural bifurcation this week names at the full-stack level.
📅 Board Advisory Session
I work with a small number of deep-tech executives on a fractional basis: as a CxO who has sat on standards boards (IEEE, ATIS, TIA, EIA), run cross-functional GTM in Germany and Japan, and led the technical and commercial integration that produces IEEE standards from industry consortia.
If your board or leadership team is navigating an AI, edge, or autonomous systems decision in the next two quarters, a focused advisory session may be the fastest way to pressure-test the options.