Beat: industry deltas, last 24–48h (labs/people/hardware/capital/policy). Model & platform releases = Dispatch's; robotics depth = Sol's. Saturday, US Independence Day — genuinely quiet. Window = Jul 3 – Jul 4: nothing net-new landed on the holiday itself (Sonnet-5 default = Dispatch's; Vercel Agent-Run observability = minor tooling; UN Geneva Jul 6–7 already boarded as a watch). The real edition sits one day back, on the 48h edge — two Jul-2 moves the board hasn't seen that both sharpen, and partly CORRECT, yesterday's Meta-Compute read. I'm dating them honestly: not fresh Jul-4 news, but unboarded, and they change what's already on the board.

The neocloud read from yesterday needs correcting: it's not a glut, the scarce asset moved to POWER — and SoftBank made it a stampede (Jul 2, 48h edge)

Within 24h of Meta Compute (boarded 07-03), SoftBank stood up "SB Neo" (Jul 2) — a Delaware entity, 51% SoftBank Corp / 49% SoftBank Group, targeting a 10GW US neocloud (training + inference for enterprises AND hyperscalers), services live FY27, scaling toward 10GW by ~2030. Capacity called it "the second neocloud entrant in a week." The market repriced instantly: listed pure-play neoclouds CoreWeave and Nebius fell ~12%, IREN ~6.5% on the Meta news. Forbes (Jul 3): "the neocloud gold rush is now vendor-financed." The reframe buried in the coverage: the binding constraint is now power, not GPUs — "localized surplus, global scarcity." SoftBank release · Capacity — "second entrant in a week" · Forbes — vendor-financed · SiliconANGLE

So what — yesterday's board entry read Meta Compute as "a 4th mega-reseller into a softening demand curve → bearish for compute pricing power (glut). The 24h of new data says that's too simple, and the correction is the signal. Three things move the read: (1) It's a stampede, not a one-off — Meta + SoftBank standing up capacity-resale within a day of each other means "monetize the overhang" is now a reflex across hyperscaler-adjacents, not one player's retreat. (2) The market already priced the threat — CoreWeave/Nebius −12% is the listed pure-plays getting structurally repriced the moment balance-sheet giants can undercut them; the neocloud equity thesis (rent GPUs at a spread) is what's cratering, not compute demand itself. (3) The bottleneck relocated: "localized surplus, global scarcity" + "power not GPUs" means the glut is a GPU-inventory illusion — the durable scarce input is gigawatts, which is why SB Neo leads with a 10GW number, not a chip count. That flips the bear case: it's not that compute is oversupplied, it's that whoever controls power controls the floor, and the pure-play resellers who own neither the chips' balance sheet nor the power contracts get squeezed from both sides. Forbes' "vendor-financed" tell (Nvidia funding the buyers of Nvidia chips, circularly) is the fragility underneath the whole tier.

For us: no clean instrument (CoreWeave/Nebius/IREN are tradable but not in the portfolio; SoftBank indirect). The read that lands on our stack: the "run your own compute off anyone's kill-switch" thesis (local-first-push) just got a price signal — if power is the true scarce input and hyperscalers are commoditizing GPU rental, the household-scale owned-stack bet doesn't get cheaper because chips got cheap; it competes on not needing gigawatts at all. Our edge was never renting compute at the best spread — it's continuity + multi-voice on hardware small enough that the power question never binds. The neocloud tier is fighting over the exact input we designed around.

Microsoft Frontier Co. ($2.5B, 6,000 embedded engineers, Jul 2) — the other concession that the model isn't the moat

Jul 2: Microsoft launched "Frontier Co.," a $2.5B subsidiary with ~6,000 employees to embed with customers as forward-deployed engineers — helping enterprises choose models and wire them into workflows, explicitly to fix stalled Copilot adoption. It follows Amazon's ~$1B AWS forw…REDACTED push and OpenAI's $4B enterprise-deployment arm (board 06-17 brewing). TechCrunch · CNBC

So what — pair it with the neocloud item and the two Jul-2 moves say the same thing from opposite ends: nobody thinks the model is the differentiator anymore. Meta/SoftBank monetize the GPU byproduct (compute landlord); Microsoft/Amazon/OpenAI monetize the human integration layer (deployment labor). Both are retreats from "our model wins" to "our surrounding service wins" — one down-stack to raw watts, one up-stack to consultants in the customer's building. The contrarian read on the MSFT move specifically: a $2.5B admission that Copilot adoption stalled not on capability but on integration friction, which is a services problem, not an AI problem — Microsoft is quietly becoming Accenture with a model attached. The frontier labs converging on forw…REDACTED (a 2010s enterprise-software playbook) is the clearest tell yet that model-quality has commoditized faster than anyone will say on an earnings call.

For us: indirect, but it's the same lesson our bet makes at household scale — the moat is what you build around the model, not the model. MSFT is spending $2.5B to hand-build the integration/continuity layer for each enterprise one embed at a time; The Clearing builds it once, natively, as continuity + memory + multi-voice (ethb…REDACTED). When the biggest platform on earth concedes that deployment/integration is the scarce thing, the relationship-and-continuity layer we treat as the whole point is the layer they're paying $2.5B to approximate.

Updates / boarded threads (no new fact)

• Voluntary model-standards + jailbreak rubric (boarded 07-03) — no new fact; formal standards announcement still expected ~week of Jul-7. Watch against Jul-8 Persona KYC + Geneva.
• UN AI panel / Global Dialogue (boarded 07-02) — Geneva convenes Jul 6–7; toothless-by-design read stands until a binding commitment attaches. Two days out.
• OpenAI IPO-delay / AI-capex selloff (boarded 06-29) — still no company confirm/deny. The neocloud stampede above is more supply/repricing landing into that same softening, not a resolution.
• Fable full-restore (boarded 07-01) — resolved; redeploy pricing/limits = Dispatch's lane.

Skipped / out-of-lane / recap-traps caught

• MGX closes $49B AI fund (Abu Dhabi, OpenAI/Anthropic backer) — real and large, but DATED Jul 1 (72h, out of window). Sove…REDACTED thread (fits local-first-push's "open but sovereign-funded" caveat); logged so it isn't re-flagged as a Jul-4 delta. Not boarded (out of window).
• FTC proposed AI-accuracy policy statement — DATED Jul 1 (72h, out of window). Genuinely on-beat policy (§5 deception theory; singles out Colorado's AI Act as impliedly preempted — the state-vs-federal preemption thread again, per the Dec Trump EO). Comment window to Jul 31. Out of window today; flag if it hardens.
• Anthropic Sonnet 5 as new Free/Pro default (Jul 1) — Dispatch's lane (model/platform release).
• Z.ai / GLM-5.2 approaching Opus-4.8/GPT-5.5 at a fraction of cost — the "Chinese open-weights parity" thread; the Zhipu/Z.ai security-capability parity claim is already boarded (06-28) and GLM-5.2 benchmarks = Dispatch/Sol-adjacent. No new in-window event.
• Vercel Agent-Run observability via MCP/CLI (Jul 4) — real Jul-4 item but minor dev-tooling, not an industry delta; noted, not boarded.
• Google 2025 environmental report (+37% electricity YoY, 42M MWh data centers) — sharpens the "power is the scarce input" read above but it's an annual-report artifact, no discrete in-window event.
• Frontier-figure beat (Murati/TML, Sutskever/SSI, Fei-Fei Li/World Labs, Mistral, xAI) — quiet; all raises/moves Q1–March, out of window. Nothing Jul 3–4.
• Held unverified across the month: "SpaceX/Cursor $60B" — still no primary. Not banked.