The Manifest
Port disruptions at Mundra and a possible full return to Suez threaten container capacity just as Nvidia flags chip supply shortages and the Bank of England warns that inflated AI valuations and leverage could shake financial markets.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01More disruption at Mundra Port as operators refuse to move boxesThe Loadstar
- 02UPS reorganization prioritizes global logistics over parcel deliveryFreightWaves
- 03Nvidia says supply shortages are limiting AI revenue growthSupply Chain Dive
- 04Nvidia's $3.5B MediaTek bet reveals its plan for tackling Big Tech's AI chip buildoutTechCrunch AI
- 05Bank of England chief warns that inflated AI valuations and rising leverage could trigger the next financial crisisThe Decoder
Models & Releases
4 storiesGradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio
For planners running voice interfaces like IVR or warehouse voice picking, sub-250ms latency with a decent accuracy score is the threshold that makes it usable at scale, worth testing against your current TTS vendor.
Google AI Releases TimesFM-3: A 330M Parameter Zero-Shot Foundation Model For Multivariate Time Series Forecasting
A model built to forecast multiple correlated demand signals in one pass without fine-tuning could cut the custom-model overhead for demand planning teams, worth a pilot against existing forecasting tools.
The Pentagon now has its own version of ChatGPT and Grok
Federal AI procurement now spans OpenAI, Grok and Gemini on one portal, a sign that multi-model sourcing is becoming standard practice rather than single-vendor lock-in, relevant for any large buyer negotiating enterprise AI contracts.
OpenClaw Releases OpenClaw 2.0: Guided Model Setup, 575 ms Control UI Startup, and One Trust Boundary Per Gateway
Faster setup and a clearer trust boundary per gateway matter for ops teams evaluating self-hosted agent tooling instead of another vendor subscription.
Supply Chain & Ops
5 storiesMore disruption at Mundra Port as operators refuse to move boxes
Empty-depot operators are refusing Adani's new yard-code rules, another flashpoint at India's top container gateway that could ripple into detention charges and container availability for anyone routing through Mundra.
Full Asia-Europe return to Suez would reverse head-haul demand growth
Sea-Intelligence models an 8.7% drop in head-haul demand next year if carriers fully abandon the Cape route, a reminder that capacity planning still hinges on a canal reopening nobody controls.
DHL Group completes corporate reorg
The holding company split from Deutsche Post AG formalizes what was already true operationally, worth noting for anyone tracking DHL's brand structure in RFPs and contracts.
Dollar General deploys AI across distribution centers, stores
Relex now runs ordering, lead times and supplier coordination across Dollar General's network, a real-world test case for planning software at retail scale worth watching for results, not just adoption headlines.
Williams-Sonoma to spread tariff refunds to vendors, employees
Sharing tariff refunds with vendors and staff is a rare example of a retailer passing savings downstream instead of just absorbing margin, worth watching if others follow.
Deals & Market
5 storiesNvidia's $3.5B MediaTek bet reveals its plan for tackling Big Tech's AI chip buildout
Nvidia buying into a chipmaker that also sells to hyperscalers building their own silicon is a hedge against losing exclusivity, buyers of AI infrastructure should expect more cross-investments rather than clean vendor lines.
Bank of England chief warns that inflated AI valuations and rising leverage could trigger the next financial crisis
A central bank chief flagging AI valuation and leverage risk to G20 finance ministers is worth taking seriously if your capex plans assume cheap AI compute keeps flowing.
OpenAI says its ChatGPT ad business hits a $1 billion annual run rate
Advertising inside ChatGPT reaching $1B run rate shows OpenAI diversifying beyond subscriptions, which could shape how it prices enterprise API and agent products going forward.
China's CXMT makes its first HBM3E chips, closing the AI memory gap
A domestic Chinese HBM supplier narrows the memory bottleneck that's constrained AI chip output, a slow-moving but real factor in future GPU and accelerator supply timelines.
Clipto uses AI to search terabytes of video and is now valued at $250M
A three-year-old video search startup hitting $250M on $15M ARR shows how thin the line still is between real revenue traction and valuation multiples in this market.
Research & Frontier
4 storiesKeenable AI Open-Sources NEEDLE: A Live Search Benchmark That Rebuilds Its Query Set Every Hour
Rebuilding the query set hourly stops search agents from just fetching the answer key, a useful reality check for anyone evaluating vendor claims about agentic search accuracy.
Intel report shows gap between robotics expectations and readiness
70% of manufacturing leaders expect to run robot fleets within five years but only 4 in 10 have a real strategy for mixed human-robot workforces, a gap that will show up in stalled deployments more than in press releases.
Conservative Hybrid Graph Networks for Process Systems with Learned Routing
Modeling plants where streams get throttled or bypassed on the fly, instead of assuming a fixed topology, is closer to how real process manufacturing actually behaves, worth watching if it moves beyond simulation.
DS-Lighting: Making Agent Harnesses Explicit for Data-Science Automation
Making the agent harness explicit rather than baked into a black box matters for anyone trying to reproduce or audit AI-driven data pipelines in planning or procurement systems.
Org & AI Architecture
4 storiesInstagram admits users often can't tell AI profiles from real people
Meta throttling reach for undisclosed AI profiles because people genuinely can't tell the difference is a preview of the labeling and trust problems any organization publishing AI-generated content will eventually face.
OpenAI starts charging some customers only when its AI actually works
Charging only when a task completes shifts risk onto the vendor and forces a harder conversation about who gets credit, or blame, when an agent underperforms, worth understanding before signing outcome-based contracts.
North America must 'join forces' to compete as global trade fragments, expert says
The pitch for deeper US-Mexico-Canada integration on infrastructure and workforce is really about surviving a more protectionist trade environment, and automation is squarely part of that workforce debate.
ChatGPT and Reddit now face EU's toughest online safety rules
ChatGPT now carries the same Digital Services Act obligations as major search engines and platforms, meaning risk assessments and transparency reports are coming whether or not enterprise users notice.