The Pulse™ | Week 27: Purpose-built takes the lead
Five signals from the fog. Coverage window: Mon June 22 to June 26, 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™
Two trade shows and one chip launch told the same story this week. From Chicago to Shanghai, the industry stopped celebrating general-purpose compute and started building for the specific job.
OpenAI taped out its first inference chip in nine months.
Huawei brought purpose-built Upper 6 GHz radio silicon to MWC Shanghai.
ABB shipped a physical-AI toolchain for factory robots.
A Korean startup licensed RISC-V cores for a neuromorphic edge processor.
The pattern across all five verticals: purpose-built is overtaking general-purpose, and the companies that own the specialized layer set the terms.
Share this: The 2026 compute story is no longer about more horsepower. It is about silicon and robots built for one job, owned by the company that defines the job.
🏭 INDUSTRIAL IOT: Physical AI takes over the factory floor
AUTOMATE 2026 filled McCormick Place in Chicago from June 22 to 25, the largest edition in the show’s 50-year run, with more than 50,000 attendees and 1,000 exhibitors. ABB Robotics debuted its Physical AI Toolchain, Kawasaki premiered an 8-DOF physical-AI robot, and an NVIDIA-sponsored Humanoid Robot Pavilion anchored the floor.
ABB introduced what it calls industry-ready physical AI; the show’s central concept moved robots from “program every step” to “show the robot what to do.”
Training by demonstration shortens deployment from months of integration to days of teaching.
For plant leaders: the robot is no longer the asset; the training pipeline is.
The 30-to-90-day signal: expect every major automation vendor to ship a physical-AI training layer and expect integration revenue to shift toward data and demonstration.
📡 TELECOM: Shanghai bets the next decade on Upper 6 GHz
MWC Shanghai 2026 ran June 24 to 26 and drew 37,300 attendees from 143 countries. Huawei unveiled a complete Upper 6 GHz hardware stack, ZTE showed the first 256-element U6G prototype, and China Mobile with Huawei swept the inaugural GLOMO awards for AI-powered networks.
China Mobile is the lead rapporteur for the first 6G scenario and demand project within 3GPP and has launched a 6G architecture verification satellite.
Upper 6 GHz is the spectrum battleground because it pairs 5G coverage with the capacity AI-native networks will need.
For carriers: spectrum positioning in Upper 6 GHz now shapes 6G economics later.
The 30-to-90-day signal: watch the 3GPP timeline and any Western operator response to China’s Upper 6 GHz hardware lead.
⚡ EDGE COMPUTING: Neuromorphic silicon reaches for the edge
On June 26, South Korean startup EdgeAI licensed Andes Technology RISC-V CPU cores for a neuromorphic system-on-chip built for ultra-low-power, event-driven inference. The design targets industrial IoT, smart infrastructure, intelligent video, and mobility.
The architecture computes only when an event changes, a sharp departure from always-on inference.
Event-driven silicon attacks the edge’s hardest constraint: power per inference.
For edge architects: evaluate neuromorphic and event-driven parts before standardizing on always-on NPUs.
This is the kind of distributed, power-aware design the OpenFog reference architecture, later IEEE 1934, was written to anticipate.
The 30-to-90-day signal: more RISC-V plus neuromorphic licensing as the edge fragments into purpose-built parts.
🚗 AUTONOMOUS SYSTEMS: Waymo plants a flag in Europe
On June 25, Bloomberg reported Waymo registered a German entity to prepare for autonomous ride-hailing in Europe. The Alphabet unit runs roughly 250,000 paid trips a week across five US cities and is targeting one million a week by year-end.
Europe means a new regulatory regime, not just a new market; type approval and liability frameworks differ from US state rules.
The German entity is a multi-year setup, not a launch; it signals intent and starts the clock.
For mobility and insurance leaders: track which European framework Waymo files under first.
Eight years convening OEMs and Tier 1 suppliers at the AVCC taught me that the regulatory venue, not the vehicle, decides who scales first.
The 30-to-90-day signal: watch for Waymo’s first European city filing and the EU type-approval pathway it chooses.
🤖 ARTIFICIAL INTELLIGENCE: OpenAI builds its own inference chip
On June 24, OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom chip, a reticle-sized inference ASIC taped out in roughly nine months with help from OpenAI’s own models. It is the first piece of a partnership covering 10 gigawatts of custom accelerators through 2029.
Jalapeño is inference-only and claims better performance-per-watt than current state-of-the-art parts.
Owning the inference chip lets OpenAI tune silicon to its models and detach from general-purpose GPU supply.
For technology buyers: custom inference silicon is now a competitive variable, not a hyperscaler curiosity.
The 30-to-90-day signal: expect more model owners to commission inference ASICs, and watch the GPU incumbents respond on price and roadmap.
🔥 3 Non-Obvious Takeaways
1. The moat moved from compute to the job the compute does
Jalapeño, Huawei’s Upper 6 GHz stack, and EdgeAI’s neuromorphic part are not faster general-purpose chips; they are silicon shaped to a specific workload. The advantage now belongs to whoever defines the workload, not whoever rents the most horsepower. General-purpose buyers will pay a specialization tax.
2. Physical AI and custom silicon are the same bet in two forms
ABB’s training-by-demonstration and OpenAI’s inference ASIC both replace a general tool with a purpose-built one. One specializes in the robot; the other specializes in the chip. Both shift value from the hardware to the party that owns the training data or the model that defines the part.
3. The next telecom and AV races are decided in standards rooms, not showrooms
China Mobile’s 3GPP rapporteur role and Waymo’s choice of European regulatory venue are the real moves this week. Spectrum standards and type-approval frameworks set the terms long before a product ships. The company writing the standard is positioning to own the market.
🗺️ The arc so far
This issue extends a multi-week thread across The Fog Signal™ on who owns the specialized layer of the AI stack.
The Sextant™ (May 6): “The Hidden Bottleneck in AI Compute.” Named compute capacity as the binding constraint on the AI buildout.
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 26 (June 23): “Capital picks the layer.” Markets began repricing the stack by layer.
The Pulse™ Week 24 (June 9): “The AI Stack Goes Institutional.” Marked the shift from capacity to governance and standardization.
This issue, Week 27 (June 30): OpenAI’s Jalapeño confirms the inference-ASIC fork is now real silicon, and purpose-built becomes the moat.
🧭 Where to start
If the shift from general-purpose to purpose-built is a live decision for your silicon, network, or automation road map, the Strategic Review Session is the right starting point.
❓ Question for you
Which of your four critical verticals, IIoT, Telecom, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting the shift from general-purpose to purpose-built infrastructure?
If you are not tracking specialization signals like custom-silicon commitments, spectrum standardization roles, and physical-AI training pipelines, you are making 2026 procurement calls without the leading indicators that separate early movers from laggards over the next two quarters.
This is a sharp and well-curated set of signals — the throughline you’ve identified (from general-purpose compute to purpose-defined compute) is the most important structural shift happening in silicon right now, and it’s playing out at every scale simultaneously.
The Jalapeño chip point deserves more attention than it’s getting. An inference-only ASIC taped out in nine months is a signal about organizational velocity as much as technical capability. What Broadcom and OpenAI demonstrated isn’t just a chip — it’s a new tempo for silicon development that breaks the traditional 3-5 year design cycle. That has cascading implications for everyone who sources compute from general-purpose vendors.
The Korean RISC-V neuromorphic SoC signal is the one I’m watching most carefully from an edge AI perspective. Event-driven compute — where the silicon only activates on meaningful state changes rather than continuously polling — is the architecture that makes always-on edge AI viable for battery-powered and energy-harvesting IoT devices. You can’t run even a small language model on a device that has a 3V coin cell battery unless the idle power profile is essentially zero, and event-driven neuromorphic designs are the closest thing the industry has to that today.
The Huawei Upper 6 GHz / 6G framing matters for edge AI in a specific way: the connectivity layer and the inference layer are converging. As eSIM-equipped edge devices gain reliable low-latency 6G access, the tradeoff between local inference and cloud offload becomes dynamic rather than fixed at design time. Purpose-built silicon for these devices needs to be designed with that hybrid in mind from day one — not bolted on later.
The moat observation at the end is exactly right. The entity that defines the job wins the value. Great edition.