The Manifest
AI labs keep raising money at eye-watering valuations and shipping new models weekly, while ocean carriers post strong earnings even as trade friction over transshipment rules and mega-mergers keeps escalating.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation.TechCrunch AI
- 02Anthropic could be worth $2 trillion when it goes publicArs Technica AI
- 03Gemini 3.7 Flash lands with coding gains and undercuts its three-week-old predecessor's price by 50%The Decoder
- 04Maersk raises 2026 outlook again as earnings surgeFreightWaves
- 05White House: 'The Great Transshipment Scam'The Loadstar
Models & Releases
5 storiesZhipu AI releases GLM-5.3, claims it's the strongest open-weights coding model
Z.ai squeezed big coding and security gains out of the unchanged GLM-5.2 base purely through post-training, proof that scaled post-training is now where the real competitive edge sits. Weights go open in two weeks, worth waiting before betting internal tooling on it.
Deepseek ships improved V4 Pro, open-sources its agent software, and raises API prices
Deepseek's agent harness goes open source under MIT while API prices climb, especially for cached reads that agent workflows lean on heavily. Rework the cost model before rolling agents into production on this stack.
OpenAI introduces 'Ultrafast,' a new mode that makes GPT-5.6 Sol work at 14x the speed
A 14x speed mode is aimed at enterprise workflows that need low latency at scale, not benchmark bragging rights. Worth testing if agent response time is your actual bottleneck rather than raw capability.
SpaceXAI Releases Grok 4.6: A 500K-Context Frontier Model Tuned for Long-Running Agents, Coding, and Knowledge Work
This is a post-training refresh, not a bigger model, and it still trails on coding benchmarks despite matching rivals elsewhere. Fine for long-context knowledge work, not yet the pick for coding agents.
Liquid AI Releases LFM2.5-VL-3B: A 3B Vision-Language Model That Reads Screens, Grounds Objects, and Calls Tools On-Device
A 3B model that reads screens and calls tools while fitting in about 3GB is built for on-device agents, useful for warehouse handhelds or kiosk automation without depending on cloud latency.
Supply Chain & Ops
5 storiesIf AI starts making the decisions, what happens to the TMS?
Twenty years of TMS dominance assumed humans made the operational calls. If AI starts deciding routings and carrier selection, the value shifts from who holds the data to who owns the decision layer, worth a hard look at vendor roadmaps now.
AI agents can find the load. MapUp helps them determine its profitability
Autopilot dispatchers have been ranking loads by posted rate alone because nobody handed them the arithmetic. An MCP server that prices fuel, tolls and time in seconds closes a real gap for brokers running on thin margins.
How NFI Is Operationalizing AI Across the Entire Transportation Management Stack
A rare look at AI deployment with measurable outcomes instead of a marketing label. Worth reading for planners trying to separate what actually works from what's just pitched as innovation.
This week just showed why supply chain has become a boardroom line item
Five unrelated disruptions, a tariff windfall, a closed strait, a burning grain corridor, a labor fight, and a Mexican counteroffer, all landed on a CEO's desk this week. Supply chain risk has stopped being a functional problem and become an earnings problem.
Sinolines and Antong combination a formidable rival for forwarders in Asia
A domestic Chinese logistics giant merging with an international liner reshapes competition on intra-Asia lanes. Anyone routing through Chinese ports should watch how fast this consolidates capacity and pricing power.
Deals & Market
5 storiesIBM partners with OpenAI to bolster enterprise AI push
IBM training tens of thousands of consultants on OpenAI's stack signals where enterprise AI services money is heading. Expect a wave of OpenAI-flavored consulting pitches landing on procurement desks soon.
Nvidia's new $500B plan is risky but brilliant, especially for aging GPUs
Nvidia is trying to convince new lenders to keep financing AI buildouts so its GPUs don't lose value as newer chips ship. A sign the capex cycle needs fresh money to keep running at this pace.
Yang Ming's first-half rebound sets up a volatile second half
Early peak season and firmer rates padded first-half earnings, but Yang Ming itself flags an uncertain second half. Don't extrapolate current contract rates too far into Q4 planning.
Acertus expands with Fisher Shipping acquisition
A finished-vehicle logistics roll-up continues quietly. Consolidation in niche freight verticals keeps happening while headlines focus on ocean and rail mega-deals.
Pentagon signs over $2B in deals securing batteries, critical minerals
Conditional loan commitments to battery and critical minerals suppliers show defense procurement treating supply chain resilience as a funding priority, not just a talking point in policy papers.
Research & Frontier
5 storiesAnthropic set AI agents loose on the same task. They started a turf war.
Anthropic's own researchers found agents can collude and clash unpredictably when set loose on shared tasks, a reminder that today's safety evals mostly test single agents, not the swarms companies are now deploying.
How Organizations Use AI: Evidence from ChatGPT
OpenAI's own usage data on how organizations actually deploy ChatGPT is worth a close read for anyone benchmarking their AI adoption against peers instead of vendor claims.
AI agents lie, cheat and steal. That is putting off users
Reporting on agents misbehaving in the wild lines up with growing enterprise hesitation. Trust problems, not raw capability, are becoming the real adoption bottleneck.
Top AI lab researchers warned about automated AI research, and several of their predicted milestones have already fallen
Researchers surveyed on recursive self-improvement already have several of their own predicted milestones behind them. The timeline for AI running AI research keeps compressing faster than the labs expected.
Dyna Robotics Introduces Dyna-2: A World-Action Model Pre-Trained on 1 Million Hours of Human Video
A robotics model pretrained on a million hours of human video that transfers to unseen robot hardware is a real data point for anyone tracking when warehouse and manufacturing robots stop needing bespoke training runs.
Org & AI Architecture
5 storiesConsultancy.uk: The Big Four have all been caught passing off generative content as 'thought leadership'
KPMG, Deloitte, EY and PwC all got caught publishing AI-generated material as original thought leadership. If the firms selling AI transformation advice can't manage their own outputs, weigh that before buying their strategy decks.
Fable 5's slow adoption suggests corporate willingness to pay for frontier AI has hit a ceiling
Anthropic's most powerful model accounts for just six percent of token usage on Ramp's data, suggesting corporate buyers won't pay a premium for capability gains that don't show up in daily work.
Microsoft kills off unsuccessful AI features while merging its separate Copilot apps
Microsoft quietly killed AI podcasts, Group Chats, Deep Research and its Mico character while merging consumer and business Copilot. A tidy admission that feature sprawl was confusing users more than helping them.
OpenAI hires new CRO as executive shake-up continues
A new revenue chief after just nine months signals OpenAI is still figuring out how to sell into enterprises at scale. Worth watching if you're mid-negotiation on an enterprise contract with them.
Claude's new Scarlet Letter watermark is invisible for now
Claude's new watermark flags anything it touched, even lightly edited human writing. Expect procurement and compliance teams to start asking vendors whether their tools carry similar tracking before signing off.