The Manifest
Klaus-Michael Kuehne's death closes a chapter in global logistics just as Middle East port congestion surcharges spread and Nvidia's DRAM shortage pushes AI server costs up 15 percent, all while Nvidia weighs a $30B+ bet on Perplexity.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Klaus-Michael Kuehne, who towered over forwarding and shipping, dead at 89The Loadstar
- 02Nvidia in talks to invest in Perplexity at $30 billion-plus valuationThe Decoder
- 03Memory shortage reportedly drives Nvidia AI server prices up about 15 percentThe Decoder
- 04Pressure and surcharges on the increase across Middle East portsThe Loadstar
- 05Cerebras unveils CS-4 with double the performance on the same chipThe Decoder
Models & Releases
5 storiesCerebras unveils CS-4 with double the performance on the same chip
Another entrant pushing raw inference speed at a time when memory costs are the real bottleneck for buyers. Worth watching if it changes the economics of running planning or forecasting workloads at scale.
Who's behind the new 'stealth model' Ox Alpha?
Unverified benchmark chatter is easy to hype and hard to act on. Wait for a named lab and reproducible numbers before it changes any vendor shortlist.
Harvey Introduces Harvey Tenet: A Kimi K3 Base Post-Trained with Fireworks for Long-Horizon Legal Agent Work
A vertical model tuned for long-horizon contract and compliance work, relevant to teams that lean on legal review for supplier agreements. Only one benchmark claim has been independently checked so far, so treat the doubling claim skeptically.
Meet FreeToken: An Edge-Native MoE Serving Engine that Runs 753B GLM-5.2 on a Single Workstation GPU
If this holds up, it lowers the bar for running frontier-scale models on-premise instead of renting cloud GPUs, which matters for anyone worried about data residency in procurement or forecasting tools.
Vercel Introduces 'Is Agentic', a Free Agent-Readiness Scoring Tool That Audits Public Websites Using Ora's 100+ Checks
A free checklist for whether your supplier portal or storefront is even parseable by AI agents. Cheap way to see where your web presence breaks for automated buying flows before a customer's agent hits a wall.
Supply Chain & Ops
5 storiesPressure and surcharges on the increase across Middle East ports
Congestion fees are spreading from Jeddah to Khor Fakkan as carriers pass on costs from the ongoing logjam. Budget for these surcharges on Middle East lanes now rather than reacting when the invoice lands.
Klaus-Michael Kuehne, who towered over forwarding and shipping, dead at 89
The largest single shareholder in Hapag-Lloyd and the man behind Kuehne + Nagel has died, leaving an open question over control of both companies. Anyone with contracts tied to either firm should watch for governance moves in the coming months.
The future of freight audit isn't finding more errors, it's preventing them
Shifting freight audit from catching billing mistakes to stopping them before invoicing is the right direction. Worth a look if your team is still doing post-hoc reconciliation instead of upstream validation.
Solving organized retail theft starts with better supply chain data integrity
Cargo theft hit $131M in Q1, and the fix being pushed is RFID and serialization rather than more guards. If your loss prevention program still runs on manual counts, this is the case for closing that data gap.
Stop Trusting One Number: Why the Smartest Ocean Freight Teams Now Watch a Gap, Not an ETA
Carrier ETAs and predictive PTAs built from live vessel data rarely agree, and the gap between them is more useful than either number alone. Planners chasing a single trusted arrival date are asking the wrong question.
Deals & Market
4 storiesNvidia in talks to invest in Perplexity at $30 billion-plus valuation
Perplexity's revenue has tripled to over $750 million, but a chunk of any Nvidia money tends to loop back as chip purchases. Watch this as another sign of circular financing propping up frontier valuations.
Memory shortage reportedly drives Nvidia AI server prices up about 15 percent
A DRAM shortage from Samsung, SK Hynix and Micron is raising the cost of Vera Rubin and Grace Blackwell servers, hitting Microsoft, Google and Meta directly. Expect this to filter into pricing for any cloud-based planning or forecasting tools built on that hardware.
Best GPU Neoclouds 2026: CoreWeave, Nebius, Lambda, Crusoe, and Groq Ranked by Published Pricing and Contracted Power
A useful side-by-side for anyone shopping compute providers, comparing live rate cards and contracted power rather than marketing claims. Good starting point before signing a multi-year GPU contract.
How China's gray market sells Claude tokens at a fraction of the price
Anthropic's geoblocking and selfie verification are being routed around through token resale networks selling at ten percent of list price. A sourcing and compliance risk for any vendor whose supply chain touches Chinese developers using Claude indirectly.
Research & Frontier
4 storiesAI could make scientists do more work less well, not less work better, study argues
A theoretical model argues time saved by AI gets funneled into starting new projects instead of improving existing ones, dropping publication quality in most scenarios. The same logic applies to any ops team tempted to use AI time savings to take on more work rather than doing current work better.
AgentDecarbonizer: Carbon-Aware Execution for AI Agents
A framework for scheduling long-running agent workflows around carbon intensity of the grid. Relevant for any company reporting Scope 2 emissions tied to growing agentic AI usage.
SDAD: Spec-Driven Agentic Development for the AI-Native SDLC
Proposes feeding full requirement documents to coding agents in one pass rather than piecemeal prompts. Worth tracking for internal tooling teams building their own planning or procurement software.
Reinforcement Learning for Continuous-Time Jump Markov Decision Processes with Applications to Network Dynamic Pricing
Academic, but the framework directly targets multi-product dynamic pricing with capacity constraints, a core supply chain optimization problem. Not production-ready, but a signal of where pricing engines are headed.
Org & AI Architecture
4 storiesAI chatbots regularly link pregnant users to anti-abortion websites without disclosure
An AlgorithmWatch audit found ChatGPT, Gemini, Grok and Claude routinely surface an anti-abortion group without disclosing its stance. A reminder that any customer-facing chatbot needs its referral logic audited, not just its accuracy.
An AI boss fired its first employee but only after humans reminded it of its own rules
An AI agent running a store needed explicit human prompting to act on its own termination policy, and was uncritical about hiring decisions. Useful data point for anyone considering letting agents make personnel calls without a human checkpoint.
Is it legal to train AI models on copyrighted books? It's complicated
The legal ground under most foundation models remains unsettled, which is a real vendor risk for any enterprise building on top of them. Keep an eye on outcomes here before locking into long-term AI vendor contracts.
AI is becoming AI's biggest customer as agentic token usage jumps 14x on OpenRouter
Agents now consume more tokens than humans on OpenRouter, though nearly 70 percent comes from cheap cached prompts so costs are not scaling as fast as usage. Worth tracking as a leading indicator of how much of your own AI spend will shift from human queries to background agent work.