The Manifest
Alibaba and Moonshot escalate the open-weight model race just as Prologis presses its $18.2bn bid for warehouse landlord Segro and the DOJ stands up a permanent unit to criminally prosecute customs and trade fraud.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Segro rejects third sweetened proposal from Prologis at $18.2bnThe Loadstar
- 02The cost of non-compliance just escalated from 'line item to prison sentence'The Loadstar
- 03Alibaba Previews Qwen3.8-Max, a 2.4 Trillion-Parameter Multimodal Model, Days After Moonshot's Kimi K3 Open-Weight LaunchMarkTechPost
- 04Moonshot pauses new Kimi K3 subscriptions after GPU demand maxes out in 48 hoursThe Decoder
- 05What to watch for after Jensen Huang's Japan visitTechCrunch AI
Models & Releases
4 storiesMoonshot pauses new Kimi K3 subscriptions after GPU demand maxes out in 48 hours
Demand outstripped compute in two days, a sign open-weight Chinese models are pulling real enterprise traffic, not just curious downloads. Worth watching if you're relying on any vendor whose inference capacity can hit a wall this fast.
Alibaba Previews Qwen3.8-Max, a 2.4 Trillion-Parameter Multimodal Model, Days After Moonshot's Kimi K3 Open-Weight Launch
No benchmarks, pricing, or model card yet, so treat this as a preview, not a reason to touch your model shortlist. Revisit once Alibaba publishes actual numbers.
Moonshot's Kimi K3 outperforms Fable 5 in frontend code but lags far behind in complex math
Pick a model by task, not by headline ranking. Kimi K3 is a solid fit for UI and frontend generation but weak for anything needing rigorous quantitative reasoning like forecasting or optimization work.
Feyn AI Releases SQRL, a Text-to-SQL Model Family That Inspects the Database Before Writing a Query
A model that checks schema before querying cuts the error class that breaks reporting dashboards and forecast pulls. The distilled 4B and 9B versions make this deployable on internal data without a cloud dependency.
Supply Chain & Ops
4 storiesPoland wary of China ecommerce as it eyes role as Europe's new cargo hub
A new hub chasing volume without vetting the traffic source is a bet on cheap parcel growth overwhelming capacity discipline. Worth watching if your network routes through Central Europe.
The cost of non-compliance just escalated from 'line item to prison sentence'
DOJ's new enforcement unit turns tariff evasion and forced labor violations into criminal exposure, not just fines. Compliance teams should re-audit customs documentation now, before an investigation forces the issue.
French customs shake-up raises the bar rather than closing the door
Regime 42's abolition is pushing UK shippers toward DAP terms, adding friction and cost to EU entry via France. Rework your Incoterms defaults before volume ramps for Q4.
The Loadstar Explainer Podcast: El Nino and its impact on shipping
Weather variability is becoming a routine planning input for ocean carriers, not a once-a-decade disruption. If your network model doesn't already flag El Nino scenarios, this is the year to add them.
Deals & Market
3 storiesSegro rejects third sweetened proposal from Prologis at $18.2bn
A rejected $18bn bid for one of Europe's biggest warehouse landlords signals how much logistics real estate is worth right now. Expect either a higher offer or a walk-away, and warehouse rents likely keep climbing either way.
What to watch for after Jensen Huang's Japan visit
Chip and compute deals spanning Japan's tech ecosystem hint at where the next wave of AI infrastructure capacity lands. Relevant for anyone planning compute-dependent tools two years out.
Can an Apple lawsuit derail OpenAI's hardware plans?
A legal fight over hardware ambitions is a reminder that vendor roadmaps you're counting on can stall for reasons unrelated to the product itself.
Research & Frontier
4 storiesGoogle Deepmind argues video generators already contain the world models computer vision has been missing
If video generators double as depth and segmentation engines, expect vision tooling for inspection and quality control to get cheaper and less data-hungry. Early research, but worth tracking for any computer-vision-heavy QA use case.
Perplexity AI Releases WANDR: An Open Benchmark Evaluating Research Agents That Must Search Wide And Deep
Top research agents score under 40 percent soft F1 on exhaustive-search tasks. A useful reality check before trusting an agent to compile a complete supplier or competitor list unsupervised.
AI text detectors struggle when language models mimic an author's style
Detectors missed nearly half of style-mimicked scientific text in testing. Anyone using AI-detection tools to vet vendor reports or research submissions should treat the results as a weak signal, not proof.
AI chatbots reading X-rays can be dangerously confident even when they're wrong
Overconfidence on inputs a model shouldn't judge alone is the same failure mode showing up in defect detection and quality inspection pilots. Calibration matters more than raw accuracy for any high-stakes automated review step.
Org & AI Architecture
2 stories'Odyssey' director Christopher Nolan calls AI an obvious 'Trojan horse'
Another prominent skeptic voice in the ongoing AI adoption debate. Worth noting as a data point on public sentiment, not a reason to change any deployment plan.
Nonprofit Current AI is racing to build the World Wide Web of AI, free for all
An open, non-commercial AI infrastructure push is a hedge for organizations wary of depending on a handful of frontier labs. Early stage, but a name worth knowing.