The Pulse™ | Week 28: Breaking the lock-in
Five signals from the fog. Coverage window: Mon June 29 to July 3, 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™
Five different players moved this week to break someone else’s lock and own their own layer.
The Pentagon backed open-source code to crack the Ericsson and Nokia grip on radio networks.
Palantir and NVIDIA put open models inside air-gapped government networks with full weight ownership.
Qualcomm moved to own a cross-accelerator software stack that routes around CUDA.
Honeywell finished carving itself down to a pure-play automation company.
California asserted direct authority over the autonomy rulebook.
The thread is ownership: who controls the stack, the spectrum, the model, and the rules.
Share this: The scarce resource in deep tech right now is not more compute; it is the freedom to switch vendors, and this week five players spent capital to buy it back.
🏭 INDUSTRIAL IOT: Honeywell becomes a pure-play automation company
Honeywell completed the spin-off of its Aerospace business and executed a 1-for-2 reverse stock split effective June 29, leaving Honeywell Technologies as a focused industrial automation and building-controls company. The old conglomerate is now three separate public companies.
Honeywell Technologies emerges, focused on the automation and OT layers; Aerospace and Solstice now trade separately.
A pure-play automation major is easier to value and faster to redirect capital toward industrial AI.
Reassess your automation vendor roadmaps; a newly focused Honeywell will push harder on controls and industrial AI.
Expect a sharper product cadence from Honeywell, no longer competing with its jet-engine division for capital.
📡 TELECOMMUNICATIONS: The Pentagon bets open-source cracks the RAN duopoly
The US Department of War’s FutureG program is backing OCUDU, a Linux Foundation open-source effort to replace the proprietary software inside the radio access network. The goal, in program director Tom Rondeau’s framing, is to make OCUDU the Linux of RAN and loosen the grip Ericsson and Nokia have held on Western networks since 3G.
I have watched this play run before. When Cisco, Intel, Microsoft, Dell, Arm, and Princeton co-founded the OpenFog Consortium, the durable win was contributing the reference architecture to the IEEE, where it became IEEE 1934, rather than letting one vendor own it. Open source is the sharper version of the same move.
OCUDU lists 47 committed organizations, including Nokia and Verizon, as well as the Linux Foundation.
Open-source code, not just open interfaces, is the lever now aimed at the vendor software moat.
Carriers should model a RAN roadmap in which core software is community-owned rather than vendor-licensed.
The duopoly will not be broken this year, but the software-ownership question is now on every carrier’s agenda.
⚡ EDGE COMPUTING: Inference keeps moving onto the device
Local inference gathered force. AMD and PC makers are pushing Copilot workloads onto on-device NPUs, and Qualcomm’s late-June agreement to acquire Modular adds a software stack that runs models across CPU, GPU, NPU, and custom ASIC without rewriting code for each accelerator.
Modular’s stack is a direct challenge to NVIDIA’s CUDA lock-in across data center and edge environments.
A portable software layer lowers the cost of deploying inference where the data actually sits.
Pilot local inference for latency-sensitive or privacy-bound workloads before renewing cloud inference contracts.
The economics keep tilting toward inference that runs at the edge, not only in the hyperscale cloud.
🚗 AUTONOMOUS SYSTEMS: California puts teeth in the AV rulebook
California’s revised autonomous-vehicle rules took effect July 1. They lift the ban on driverless vehicles over 10,001 pounds, opening the door to autonomous freight, permit medium-duty AV transit, and add enforcement: a roadside Notice of AV Noncompliance served on manufacturers with a 72-hour response window, geofencing directives, and a 30-second first-responder standard.
In eight years leading the Autonomous Vehicle Computing Consortium, I watched the governance layer, not the compute, decide which programs scaled. Rules like these are where dense-city autonomy is won or lost.
Driverless freight moves from pilot to permitted in the largest US state market.
Enforcement now reaches the manufacturer directly, not only the operator or the vehicle.
Fleet and logistics leaders should revisit autonomous-freight timelines under the new California framework.
Regulators are writing the governance layer in real time; the rules, not the sensors, will gate scale.
🤖 ARTIFICIAL INTELLIGENCE: Sovereign AI gets a reference design
Palantir and NVIDIA announced on June 29 an engine to run NVIDIA’s Nemotron open models inside air-gapped and classified government networks. Agencies can train on their own data and keep full ownership of the resulting model weights. Alex Karp framed the offering explicitly in contrast to the weaker IP protection at OpenAI and Anthropic.
Nemotron open models run in classified, air-gapped environments with full customer weight ownership.
Sovereign, on-premises AI now has a named reference design, not just a slogan.
Regulated enterprises should ask whether they own their model weights or merely rent access to them.
Ownership of model weights becomes a procurement question, not a philosophical one, over the next two quarters.
🔥 3 Non-Obvious Takeaways
1. Lock-in is the contested asset this week, not compute
Three of the five signals attack a different vendor lock: OCUDU targets the Ericsson and Nokia RAN software moat, Qualcomm’s Modular deal targets CUDA, and Palantir with NVIDIA targets the closed-model cloud. The scarce resource executives are fighting over is not more silicon; it is the freedom to switch.
2. Sovereignty is descending the stack
A year ago, sovereign AI meant a national cloud. This week it meant open model weights in an air-gapped room, community-owned RAN software, and inference on the device in your hand. Control is migrating from the data center down to spectrum and silicon, and whoever holds the lower layer sets the terms above it.
3. The pure-play returns as an ownership strategy
Honeywell finished unbundling into a focused automation company the same week California asserted sole authorship of the AV rulebook. Both are bets that owning one layer cleanly beats straddling several. For boards, the question shifts from what do we do to which layer do we intend to own.
🗺️ The arc so far
This issue extends the arc from who sets the standards to who owns the layer.
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 25 (June 16): “Owning the layer.” Reframed the contest as control of a defensible layer over raw performance.
The Pulse™ Week 27 (June 30): “Purpose-built takes the lead.” Showed purpose-built silicon and physical AI overtaking general-purpose compute.
This issue, Week 28 (July 7): five players spend capital to break someone’s lock-in and own their layer outright.
🧭 Where to start
If deciding which layers of your stack you truly own, and which you rent from a vendor who can change the terms, is a live question for your leadership team, the Introductory Session is the right place to begin.
❓ Question for you
Which of your four critical verticals, IIoT, Telecom, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting where a vendor lock is quietly deepening?
If you are not tracking your vendor concentration ratio, the share of inference you run on infrastructure you own versus rent, and how open your RAN or control software actually is, you are making 2026 procurement calls without the leading indicators that separate early movers from laggards over the next two quarters.