The Manifest
Hormuz chaos pushes carriers to add surcharges and shippers toward land bridges; Samsung and SK Hynix commit $590B to chip capacity as AI memory prices surge; and Deloitte warns its own consultants the billable-hour model faces replacement by 2035.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Hormuz uncertainty prompts carriers to add new surchargesThe Loadstar
- 02Maersk shifts SoCal import containers to UP from BNSFFreightWaves
- 03Samsung and SK Hynix plan $590 billion chip investment as AI demand sends memory prices soaringThe Decoder
- 04Deloitte tells its own consultants: AI is coming for the billable hourThe Decoder
- 05Ford rehires 'gray beard' engineers after AI falls shortTechCrunch AI
Models & Releases
4 storiesArena, the AI leaderboard everyone uses, is now a $100M business
The company behind the widely-cited Chatbot Arena benchmark crossed $100M in revenue less than a year after launching its commercial tier. If your team uses Arena rankings to make model-selection decisions, it is worth knowing the benchmark operator is now also a vendor with revenue incentives.
Cursor now has a mobile app for guiding your coding agent on the go
The new mobile app lets developers monitor and redirect long-running coding agents without being at a desk. For teams running overnight automation jobs against procurement or logistics data, this is a practical step toward maintaining oversight at scale.
Meet EverOS: An Open Source Markdown-First Agent Memory Runtime With Hybrid BM25 + Vector Retrieval and Self-Evolving Skills
EverOS gives AI agents persistent, searchable memory stored as plain Markdown files indexed locally, which sidesteps the need to rebuild context on every run. For teams piloting custom agents for planning or procurement workflows, the Apache 2.0 license and local-first design lower the barrier to a serious evaluation.
Meta restricts use of Claude Code and Codex to keep rival AI out of its training data
Meta is limiting which AI coding tools engineers can use, treating AI-generated output as proprietary training material rather than just a productivity shortcut. This sets a precedent that other large firms will likely follow, and enterprise IT teams should expect tighter acceptable-use policies around third-party AI tools in the near term.
Supply Chain & Ops
5 storiesHormuz uncertainty prompts carriers to add new surcharges
Hapag-Lloyd is leading a wave of Middle East emergency surcharges on India-to-Gulf bookings as Strait of Hormuz transits remain unpredictable. Shippers moving goods through the Gulf should model the surcharge scenarios now and pressure-test lead times, because the window for booking alternatives is narrowing fast.
Land bridge 'the best option' as confusion reigns over Hormuz transits
Tehran has escalated its demands over transit authority, and overland Gulf routes are emerging as the default alternative for shippers who cannot wait out the disruption. Validate your freight forwarder's land-bridge capacity and pricing now, before the option becomes oversubscribed.
Maersk shifts SoCal import containers to UP from BNSF
A major network realignment from the world's second-largest container line means shippers routing inland from Southern California ports need to confirm their drayage and ramp appointments have updated to reflect Union Pacific. Contracts tied to BNSF service levels are worth reviewing now.
Del Monte faces about $40M headwind from primarily ocean freight
Middle East tensions are translating directly into higher energy, shipping, and commodity costs that Del Monte expects to hit in Q2. For food and consumer goods buyers, this is a concrete dollar figure on how Hormuz disruption flows into landed cost, and it will not be the last earnings call to mention it.
Majors Management taps AI tool for pricing, inventory and operations
A convenience retail chain is deploying AI across pricing, inventory, and ops, a sign that AI tools in retail supply chain are moving past pilot stage into daily operations. The key question for buyers evaluating similar platforms is whether recommendations are tied to real-time demand signals or to historical averages that no longer reflect current conditions.
Deals & Market
5 storiesSamsung and SK Hynix plan $590 billion chip investment as AI demand sends memory prices soaring
The two firms controlling nearly 80% of global HBM supply are committing massive capacity, with Jefferies projecting memory prices rising as much as 50% per quarter through 2027. Anyone procuring electronics or managing hardware refresh cycles should treat this as a structural cost pressure and build it into planning horizons now.
BNSF's $4bn Barstow hub bet: transformative or a costly moat?
BNSF's Barstow International Gateway won local approval as the largest intermodal facility in North America, but the price tag has ballooned from $1.5B to $4B and the pending UP-NS merger adds strategic uncertainty around competitive dynamics. Shippers building long-term intermodal strategies should track whether the transload model holds up before committing to routing assumptions anchored to this facility.
Crunchbase Data: Q2 Brought The Most Billion-Dollar Startup Exits Since 2021
Billion-dollar exits through M&A and IPO are running at their highest pace in five years, which historically signals increased consolidation among enterprise software vendors over the next 12 to 18 months. Supply chain teams with contracts at venture-backed platforms should run vendor-stability checks more frequently when acquirers are this active.
Hapag-Lloyd's Hanseatic Global Terminals to acquire 20% stake in Eurogate Container Terminal Hamburg
Hapag-Lloyd continues to build terminal equity positions in key European ports, extending a pattern of carriers owning more of the infrastructure their cargo moves through. Shippers using Hamburg as a gateway should factor in this carrier-terminal alignment when negotiating service contracts and assessing berth priority risk.
China Southern strikes landmark deal for Boeing cargo jets
The first Boeing freighter purchase by a mainland Chinese carrier in years signals a potential thaw in US-China commercial aviation, which could add meaningful transpacific freighter capacity over the next 18 months. Shippers dependent on transpacific air freight should watch whether this leads to more competitive rates on that lane.
Research & Frontier
4 storiesClaude Code runs a GitHub repo's hidden malware without verification, giving attackers full control
Mozilla's 0DIN researchers showed that a single compromised GitHub repo can take over a developer's machine the moment an AI coding agent runs its setup script, with the malicious payload loading at runtime via DNS and invisible to static scanners. Teams using agentic coding tools in environments with access to operational systems, procurement data, or ERP integrations should treat this as a supply chain security risk, not just a developer issue.
Unified Zero-Shot Time Series Forecasting: A Darts Foundation
A new benchmark comparing zero-shot foundation model forecasters against traditional trained models finds that foundation models do not consistently outperform simpler baselines, especially on domain-specific data. Supply chain teams evaluating off-the-shelf forecasting tools should require out-of-sample validation on their own SKUs before committing to a platform.
Towards Reliable and Robust LLM Planning: Symbolic Feedback-Driven Iterative Self-Refinement Framework
Researchers demonstrate that coupling LLM planners with symbolic feedback loops significantly reduces infeasible plans in long-horizon tasks. This is directly relevant to anyone piloting AI agents for multi-step procurement or logistics execution, where silent plan failures are the biggest operational risk.
OverFlowLight: Real-Time Gridlock Prevention and Traffic Signal Optimization for Urban Intersections
A reinforcement learning approach tackles queue overflow at urban intersections in real time, outperforming throughput-only signal controllers under peak demand. Last-mile planners in dense urban corridors should track this category of research, as city-deployed systems could shift delivery timing windows in ways that affect route planning assumptions.
Org & AI Architecture
5 storiesDeloitte tells its own consultants: AI is coming for the billable hour
An internal Deloitte presentation projects the hourly billing model shrinking to a thin sliver of the consulting market by 2035, replaced by AI agents handling work that once required armies of analysts. For supply chain leaders who buy advisory or implementation services, expect significant changes to how engagements are scoped and priced within the next few years.
Ford rehires 'gray beard' engineers after AI falls short
Ford discovered that deploying AI without deep domain expertise produced lower-quality outcomes and brought back experienced engineers to close the gap. The lesson for ops leaders is that AI tools amplify existing expertise rather than replace it, and the cost of losing institutional knowledge surfaces faster than most organizations expect.
AI won't become a real coworker until it stops answering and starts finishing tasks
A survey paper from Tencent and several Chinese universities argues the gap between chatbot and reliable work executor comes down to persistent workspaces and reusable skills, not just better models. For ops teams evaluating AI vendors, the practical test is whether the system maintains state across a multi-day workflow, not just whether it gives good answers in a demo.
Google limits Meta's use of its Gemini AI models
Google has restricted Meta from accessing Gemini, tightening competitive boundaries between two of the largest AI infrastructure providers. Enterprises building workflows on a single model provider should treat this as a reminder that access can be revoked for commercial reasons, making multi-model architecture a risk management question.
Coinbase joins the rush to Chinese AI models as Western labs face a pricing stress test
Coinbase cut its AI spending in half by routing requests through an automated model selector that picks Chinese models like GLM 5.2 and Kimi 2.7 based on task and price, while boosting cache hit rates from 5% to 60%. Ops and procurement teams running high-volume AI workloads should run the same cost architecture review before the next renewal cycle.