The Pulse™ | Week 32: The conversion test
Five signals from the fog. Coverage window: Mon July 27 to 31, 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™
For three years, the industry argued about how much capacity to build. This week the market started asking what the capacity converts into, and it priced the answer in public. Microsoft and Amazon showed conversion and were rewarded. Meta and Alphabet showed spending and were not.
The same test ran quietly in the other four verticals.
Verizon repriced a fiber network built for a different era.
Qualcomm passed a memory bill through to its customers.
A humanoid robotics company went public on operating hours rather than demonstrations.
A proposed bill moved the autonomous vehicle argument from performance claims to response protocols.
Share this: Capacity stopped being the differentiator this week. Conversion took its place: revenue per gigawatt, margin per contract, operating hours per robot, price per token.
🏭 INDUSTRIAL IOT: Humanoids get priced on operating hours
Agility Robotics opened an AI training hub in Fremont, California on July 29, ahead of a Nasdaq listing through a $2.5 billion merger with Churchill Capital Corp XI. What it is taking to public investors is not a demonstration reel. It is a service record.
Digit runs at nine customer sites including Schaeffler, GXO, Toyota Motor Manufacturing Canada and Mercado Libre.
More than 65,000 hours logged in live production, the first reliability baseline the category has offered.
Ask every humanoid vendor for hours in production, not units shipped or payload specifications.
Public markets will now hold humanoid robotics to the same uptime arithmetic that governs every other machine on the plant floor.
📡 TELECOMMUNICATIONS: Carriers reprice fiber as AI infrastructure
Verizon detailed a $1 billion dark fiber agreement with Google to connect AI data centers, disclosed on its July 24 earnings call and reported July 27. Chief Executive Dan Schulman said further deals worth multiple billions would close by year-end, with revenue layering in from 2027.
Schulman put AI Connect margins at equal to or greater than the legacy portfolio and called it a very different growth profile.
SK Telecom’s board approved SK Hyper the same day, up to KRW 750 billion through 2030 toward 15 GW of AI data center capacity by 2035.
Reopen long-haul and metro fiber terms now; contracted AI transport is repricing faster than access.
Assets built for the last era are being underwritten for this one, and the contracts run long.
⚡ EDGE COMPUTING: The memory bill arrives at the edge
Qualcomm posted fiscal third quarter revenue of $9.95 billion on July 29, above consensus, then guided the current quarter to $2.05 to $2.25 per share against roughly $2.35 expected, citing a memory supply squeeze. Chief Executive Cristiano Amon said prices will rise across the board on September 1.
Automotive revenue set a company record, up 61% year over year on rising compute content per vehicle.
Arm reported $1.289 billion for the same quarter, up 22%, on cloud and edge architecture adoption.
Put memory footprint in the 2027 edge budget model; silicon price is no longer the only variable.
When I co-founded the OpenFog Consortium and we wrote the reference architecture that became IEEE 1934, the constraint we designed around was latency. In 2026, the binding constraint at the edge is bill of materials.
🚗 AUTONOMOUS SYSTEMS: Emergency response becomes a design requirement
On July 28 in San Francisco, Representative Kevin Mullin of California unveiled the AV Emergency Response Coordination Act alongside Mayor Daniel Lurie and Fire Chief Dean Crispen. It would direct NHTSA to set minimum national emergency response standards for driverless operators.
Operators would publish first responder protocols and staff a 24-hour hotline for public officials.
Cities would gain a real-time geofencing lever to restrict robotaxi operation during active incidents.
NHTSA Administrator Jonathan Morrison’s end-of-July deadline for fixes passed without published solutions.
Across eight years leading the Autonomous Vehicle Computing Consortium, the hardest governance question was never how the vehicle sensed the world. It was defining what the vehicle owes the people working beside it. That question is now being written into statute.
🤖 ARTIFICIAL INTELLIGENCE: The market splits capex from conversion
Microsoft reported on July 29 that Azure crossed $100 billion in annual revenue and accelerated to 43% growth, with an AI run rate of $37 billion, up 123%. Shares rose more than 15%. Amazon followed on July 30 with AWS at $42.2 billion, up 37%, its fastest in eighteen quarters, and capital spending guided near $220 billion.
Meta grew revenue 28% to $60.8 billion, but profit fell 14% to $15.85 billion on infrastructure costs.
Alphabet guided 2026 capex to $195 to $205 billion and posted its first negative quarterly free cash flow in roughly two decades.
The European Commission opened its AI Gigafactories call on July 30, targeting over 30 billion euros against roughly 1 billion committed.
The four largest buyers of AI infrastructure now report on different terms. Expect supplier negotiations over the next quarter to split along the same line.
🔥 3 Non-Obvious Takeaways
1. Depreciated assets are outperforming new capex this cycle
Verizon’s long-haul fiber was built for a different business and is now the highest-margin thing it sells into AI. Agility went public on 65,000 accumulated hours, not on new hardware. The infrastructure winning in 2026 is disproportionately the infrastructure that was already paid for, because it carries no fresh depreciation against uncertain demand.
2. The memory shortage is an allocation problem before it is a supply problem
Qualcomm raises prices on September 1 because it cannot secure memory at the cost it assumed, while the hyperscalers commit roughly $725 billion to systems built from the same components. Industrial and edge buyers are now queued behind balance sheets a thousand times their size, and no standards body arbitrates that queue.
3. Regulators reached the AV industry through operations, not autonomy
The bill unveiled July 28 does not legislate perception stacks or disengagement rates. It legislates hotlines, protocols and geofences. The governance frontier moved from how the vehicle drives to how the operator answers the phone, which is precisely the function most AV programs staffed last.
🗺️ The arc so far
This issue closes a five-part argument that began with who governs the stack and ends with who gets paid for it.
The Sextant™ (May 20): “The Hidden Lock-In Beneath Inference and Physical AI.” Established that the durable advantage sits in the layer buyers cannot easily leave.
The Pulse™ Week 28 (July 7): “Breaking the lock-in.” Five players spent capital to escape vendor dependence and own their layer outright.
The Pulse™ Week 29 (July 14): “The map is the moat.” Advantage descended from technology into scarce physical position: spectrum, power, territory.
The Pulse™ Week 30 (July 21): “Owning the inference layer.” Ownership moved into the inference layer itself, confirmed by Meta’s Iris silicon.
The Pulse™ Week 31 (July 28): “The price of the layer.” Competition shifted from raw capacity to unit economics: tokens per dollar, watts per inference.
This issue, Week 32 (August 4): The market applied the price. Owning a layer now has to show conversion, quarter by quarter.
🧭 Where to start
If your board is being asked to approve capacity this quarter and cannot yet answer what that capacity converts into, the Board Advisory Session is the right starting point.
❓ Question for you
Which of your four operational verticals, Industrial IoT, Telecommunications, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting when infrastructure spend stops converting into revenue?
If you are not tracking revenue per deployed gigawatt, gross margin on contracted infrastructure, and component cost per shipped unit against a rolling forecast, you are making 2027 procurement calls without the leading indicators that separate early movers from laggards over the next two quarters.