The Manifest
Moonshot's giant new Kimi K3 model headlines a busy release day while Walmart, Kuehne+Nagel and TSMC show AI reshaping physical supply chains, and a wave of enterprise surveys plus a Hyundai plant strike highlight growing friction between AI ambition and operational trust.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Moonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M ContextMarkTechPost
- 02Fear of humanoid robots spurs human workers to strike at Hyundai auto factoryArs Technica AI
- 03Walmart bets on AI and digital twins to shape its supply chain strategySupply Chain Dive
- 04TSMC says AI demand can run to 2030, lifting the outlook for air cargoThe Loadstar
- 05Kuehne + Nagel rolls out cloud-native contract logistics solutionThe Loadstar
Models & Releases
5 storiesMoonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M Context
Full weights land July 27, and early benchmarks put it near Claude and GPT tier at a much higher price than its predecessor. Worth testing for long-document planning work, but budget for the fact that cheap Chinese frontier models may be ending.
NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB
A strong open embedding option for anyone building retrieval over supplier docs, contracts or spec sheets. The 1B distilled variant is the practical pick if you need throughput on existing GPUs rather than the top benchmark score.
Ex-OpenAI CTO Murati's Thinking Machines drops Inkling, a 975B parameter model that leads US labs but trails China
Another large open-weights entrant pitched as a fine-tuning base rather than an end product. The real story is the widening gap: US open models still trail Chinese ones on several benchmarks despite the parameter count.
Gemma 4 gets a stealth update that fixes tool calling bugs and truncated responses under the same name
If you built agent workflows on Gemma 4 and hit flaky tool calls, this quiet patch is worth re-testing before blaming your own prompts. No version bump makes it easy to miss.
German AI consortium releases Soofi S, an open 30B model that tops benchmarks
A smaller, deployable option for teams that need strong German and English performance without the compute bill of a trillion-parameter model. Good candidate for on-prem or regulated environments.
Supply Chain & Ops
5 storiesKuehne + Nagel rolls out cloud-native contract logistics solution
KN SwiftLOG moving to a cloud-native, agentic-AI-enabled platform across 1,000-plus sites is a signal that warehouse management systems are becoming the next battleground for AI vendors, not just TMS and planning tools.
Walmart bets on AI and digital twins to shape its supply chain strategy
Digital twins for routing product around conflict zones and weather disruption are moving from pilot to core strategy at scale. Smaller operators should watch which vendors Walmart validates, since that tends to set the market template.
TSMC says AI demand can run to 2030, lifting the outlook for air cargo
Semiconductors and data center gear have overtaken smartphones as air cargo's fastest-growing category. Freight buyers should expect continued capacity pressure and pricing power on lanes serving chip and server supply chains.
Renfro broadens AI tools for tighter supply chain compliance
Expanding Inspectorio use for supplier performance and regulatory tracking is a practical, unglamorous example of AI adoption that actually reduces audit and compliance headcount pressure rather than chasing a flashy use case.
Exporters demand compensation after floods disrupt Chittagong port
A week of heavy rain shut down Bangladesh's main export gateway, hitting garment and textile shippers hardest. Expect knock-on delays into already tight Asia-Europe capacity over the coming weeks.
Deals & Market
5 storiesEnergy IPOs surge as investors hunt for ways to play AI boom
Capital is flooding into power generation and grid plays as the fastest route to AI exposure. Anyone planning data center or manufacturing capacity should watch this for early signals on where new power supply is actually landing.
Apple Intelligence approved for launch in China with Alibaba and Baidu
A long-delayed regulatory approval that finally gives Apple a compliant AI story in its largest overseas market. Watch how this shapes localized AI features and data handling requirements for other Western vendors selling into China.
The AI compute gap: Enterprises are buying infrastructure faster than they can measure what it costs
A majority of enterprises plan to switch or add compute providers within the year, often within a quarter, while still lacking clear cost visibility. Anyone signing new compute contracts should push for usage transparency before committing.
US tags Brazil imports with 25% tariff
The new levy hits July 22 with carve-outs for goods already under Section 232 duties. Buyers sourcing from Brazil should audit exposure now rather than after the effective date.
How a former DeepMind researcher raised at a $300M pre-seed valuation before launching a product
A reminder that pedigree alone is commanding pre-product valuations in visual AI. Worth tracking as a signal for where investor enthusiasm is outrunning demonstrated traction.
Research & Frontier
4 storiesOpenAI Details GPT-Red: An Internal Automated Red-Teaming Model That Beat Human Red-Teamers 84% To 13% On Prompt Injection
Automated red-teaming beating human testers by that margin is a strong argument for adding adversarial AI testing to any procurement or deployment checklist involving agentic tools. Multi-turn attacks remain the weak spot to watch.
Intelligent Three Level Learning Architecture for Autonomous UAV Swarms in Search and Rescue
The hierarchical learning approach is aimed at disaster response but the coordination methods translate directly to warehouse and yard drone fleets managing inventory checks at scale.
ToolAnchor: Anchoring Counterfactual Context to Boost Agentic Tool-use Capability
Addresses a real operational headache: agents trained on a fixed toolset struggle when new tools get added without full retraining. Relevant for any team building procurement or logistics agents that need to pick up new APIs over time.
Branching Policy Optimization: Sandbox-Native Language Agent Reinforcement Learning
A more efficient training approach for agents that operate in executable environments, which matters as more supply chain tools move from chat interfaces to agents that actually execute tasks in systems.
Org & AI Architecture
5 storiesNetflix's 300 AI productions show how fast the technology is spreading through entertainment
Producing footage twice as fast at half the cost, with savings funding more content rather than shrinking budgets, is the adoption pattern to watch: AI reinvested into output volume rather than headcount cuts, at least for now.
Fear of humanoid robots spurs human workers to strike at Hyundai auto factory
Hyundai's plan to deploy 25,000 Atlas robots starting in US factories by 2028 turned into a labor flashpoint before a single unit shipped. Any manufacturer planning robot rollouts should treat workforce communication as a project workstream, not an afterthought.
The agent security gap: 54% of enterprises have already had an AI agent incident, and most still let agents share credentials
Only a third of enterprises give every agent its own scoped identity, which means most agent deployments are one bad prompt away from a real incident. Basic identity hygiene is overdue before scaling agent access further.
The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem, and most are shipping to production anyway
Half of surveyed enterprises shipped an agent that passed internal evals and then failed a customer in production. Treat internal benchmarks as a floor, not a launch gate, and build in real-world monitoring before go-live.
Linus Torvalds to critics of AI coding in Linux: 'Fork it. Or just walk away.'
A blunt data point on how deeply AI-assisted coding has already embedded itself in critical infrastructure projects. Leaders debating internal AI coding policy should note that resistance is losing this argument in the open source world too.