The Pulse™ | Week 23: Capacity meets the build queue
Five signals from the fog. Coverage window: Mon May 25 to May 29, 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™
Last week, capacity got an address. This week, every layer of the stack exposed a different waiting line.
Tesla broke ground on a dedicated Optimus factory at Giga Texas while operating 42 robotaxis in the state.
Texas turned on a public AV registry.
PwC put power-transformer lead times at up to four years; Bloomberg put only about a third of 2026 US data-center capacity actually under construction.
T-Mobile and Ericsson moved an AI-native radio scheduler toward commercial release.
The question is no longer whether you can fund capacity. It is how long the build queue is.
Share this: Capacity meets the build queue in 2026. Capital is unconstrained; transformers, fabs, factories, and operating fleets are not. The constraint is industrialization, not appetite.
🏭 INDUSTRIAL IOT: Optimus factory breaks ground at Giga Texas
On May 27, the first steel was put up at Tesla’s dedicated Optimus factory at the North Campus of Gigafactory Texas. The permit package adds over 5.2 million square feet at an estimated $5-$10 billion, with a stated long-term target of 10 million units per year and first production at the second-generation facility in summer 2027.
Tesla also moved Optimus pilot production to Fremont, freeing Giga Texas for the scaled build.
The deployable unit is the factory shift, not the press release; concrete poured today ships robots in 2027.
Evaluate humanoid programs against named lines and named hours, not fleet promises.
Expect humanoid procurement to split between operators who can name the line and the shift, and those still scoping pilots.
📡 TELECOMMUNICATIONS: AI moves into the radio scheduler
T-Mobile and Ericsson concluded a live AI-native Scheduler with Link Adaptation trial on T-Mobile’s 5G Advanced network and confirmed commercial deployment in the third quarter. The model predicts shifting wireless conditions in real time, delivering roughly 10% spectral efficiency and 15% download-speed gains without new spectrum or new radios.
AT&T tracked roughly half its wireless traffic onto Open-capable platforms by early 2026, targeting 70% by year-end.
AI shifts from monitoring the network to scheduling it; the radio layer becomes a software-defined lever for throughput.
Price the scheduler software upgrade against base-station refresh cycles, not against new spectrum.
In my career go-to-market years at Lantiq and Marvell, where qualification at NTT Labs was the gold standard, spectral efficiency was an antenna and waveform problem. It is now a model problem, and the model ships faster than the radio.
⚡ EDGE COMPUTING: Power lead times push inference outward
PwC confirmed that high-capacity transformer lead times stretched to four years; Bloomberg reported that only about a third of the 12 to 16 gigawatts of US data-center capacity targeted for 2026 were under construction. The substation is now the binding constraint, not the chip. The arithmetic favors distributed inference on owned edge sites.
Crusoe Energy began manufacturing its own switchgear to bypass utility lead times; hyperscalers are pre-ordering years out.
Inference economics rewards workloads close to data and load, not those routed back to a central region.
Map which inference workloads can move to owned edge capacity before renewing central cloud commitments.
Compute close to where data is produced is the operating basis of edge computing, and it is what the IEEE 1934 reference architecture, which came out of the OpenFog work I helped lead, was written to standardize. Grid economics now ratifies it from the other side.
🚗 AUTONOMOUS SYSTEMS: Texas turns on the AV registry
Texas Senate Bill 2807 took effect on May 28, requiring every operator or tester of automated vehicles to register with Tx-DMV and disclose fleet size and safety information. The first public registry: Waymo 577, Avride 317, Nuro 47, Tesla 42, Zoox 35. Waymo runs more than thirteen times Tesla’s fleet in the state Tesla has positioned as its commercial launch market.
First regulator-mandated public reading of US operating-fleet sizes by AV operator.
Marketing claims and operating fleets are now publicly comparable; the announcement-to-operation gap is no longer optional disclosure.
Weight TxDMV registry counts and operating-domain restrictions over coverage maps when evaluating AV partners.
Across the eight years I founded and led the AVCC, autonomy was bounded by the operating domain in front of the vehicle. Texas just made the operating fleet behind the vehicle equally subject to public records.
🤖 ARTIFICIAL INTELLIGENCE: Vera Rubin meets the Taiwan supply chain
Jensen Huang flew to TSMC on May 24 ahead of his June 1 Computex Taipei keynote, with reporting describing the Vera Rubin ramp as straining the Taiwan supply chain. Rubin is in full production with partner products in the second half of 2026; TrendForce puts the top nine cloud and AI providers at roughly $830 billion in 2026 capex. The chip side is shipping. The grid and the fab are the new pacing layers.
Microsoft is in talks to supply its Maia chip to Anthropic, further unbundling training capacity from any one provider.
Custom silicon volume is a multi-supplier story; the bottleneck shifted to packaging, HBM, transformers, and qualified power.
Test chip-allocation contracts against power-delivery dates, not chip-availability dates.
Expect the second-half narrative to move from capex figures to delivery dates: when megawatts arrive, when the substation energizes, when the chips enter racks.
🔥 3 Non-Obvious Takeaways
1. The deployment gap is now a regulated disclosure
Texas SB 2807 made AV fleet size a public number; the same week, Tesla’s permit package made its humanoid build a public number, and PwC made grid lead times a public number. Three different verticals, one pattern: the gap between marketing announcement and operating reality is being forced into the record. Board credibility now depends on which number you cite.
2. The next bottleneck is the substation, not the chip
Anthropic’s TPU commitments, NVIDIA’s Vera Rubin ramp, and Akamai’s GPU build-out across edge sites are all moving faster than the transformers and switchgear that energize them. The four-year lead time on high-capacity transformers reroutes inference toward distributed edge capacity by economic necessity, not by architectural preference.
3. AI quietly took the radio scheduler
T-Mobile and Ericsson’s commercial commitment lifts roughly ten percent of spectral efficiency out of software, not silicon. It is the first vertical where AI control loops shift unit economics without a hardware refresh. Expect the same pattern in grid dispatch, factory scheduling, and AV fleet routing inside twelve months.
🧭 Where to start
If the deployment gap and the build queue are now live decisions for your leadership team, the Board Advisory Session is the right starting point to map procurement and capital commitments against the bottlenecks that will actually pace 2026:
❓ Question for you
Which of your four critical verticals, IIoT, Telecom, Edge Computing, or Autonomous Systems, has the weakest instrumentation for detecting the deployment gap between what your vendors announce and what they actually operate?
If you are not tracking operating fleet size by vendor, energized megawatts by site, and the lead-time variance between announced and contracted equipment, you are making 2026 procurement calls without the leading indicators that will separate early movers from laggards over the next two quarters.
🗺️ The arc so far
The Sextant™ (2026-05-06): “The Hidden Bottleneck in AI Compute”. Established the substrate-as-constraint thesis.
The Pulse™ (2026-05-12): “Physical AI Goes Operational”. Moved physical AI from pilot to production line.
The Vector™ (2026-05-14): “The Inference ASIC Fork”. Architectural split between hyperscaler and edge inference.
The Pulse™ (2026-05-26): “Capacity gets an address.” Compute, coverage, autonomy re-sited as owned infrastructure.
The Sextant™ (2026-05-20): “The Hidden Lock-In Beneath Inference and Physical AI”. Board-level lock-in analysis on inference and physical AI.