The Manifest
Anthropic's slowing growth ahead of its IPO and Nvidia's European compute pledges dominate the market conversation while a sunk container ship in the Strait of Hormuz and a port-sector data leak threat keep supply chain risk front and center.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Anthropic's Q3 revenue hits $16.8B but growth slows ahead of planned November IPOThe Decoder
- 02GFS ship sinks after missile attack, second ship sustains damageThe Loadstar
- 03ASML's strong orders ease fears AI spending boom is running out of steamreuters.com
- 04Nvidia's Jensen Huang pledges $2 billion for Europe, $250 million in compute for UK safety labsThe Decoder
- 05No post-peak easing at US container gateways, despite drayage woesThe Loadstar
Models & Releases
5 storiesMeet Grok Voice 3.0: SpaceXAI Voice Agent Model Ranks #1 on Artificial Analysis
A speech-to-speech model that listens, reasons and calls tools now tops the voice benchmark at the same price point as its predecessor. Worth a look for anyone piloting voice bots in customer service or dispatch, since the cost bar for a top-tier option just didn't move.
Figma announces new features to define how agents work on its platform
Figma is formalizing how agents touch design files, plus opening code layers to beta. If your team is already using AI to turn mockups into shippable code, this narrows the gap between design and dev handoff.
Google Releases Gemini 4 Argon API With 1M Token Output Limit
A 1M output token ceiling matters for anyone generating long structured documents, like full contract redlines or multi-site compliance reports, in a single pass instead of stitching chunks together.
Liquid AI Releases LFM2.5-24B-A2B: A 2B-Active MoE for On-Device Agents
A small-active-parameter model that runs fast on consumer hardware is the kind of thing that makes offline, on-device agents in warehouses or trucks practical without a cloud dependency.
Fish Audio Releases Drama 3, a 35B Open-Weight MoE Text-to-Speech Model
Open weights and 83 languages lower the cost of multilingual voice support, which is relevant to anyone building IVR or driver-facing voice tools across a global network.
Supply Chain & Ops
5 storiesNo post-peak easing at US container gateways, despite drayage woes
Record volumes at Houston and other gateways mean drayage will stay the bottleneck heading into Q4. Planners should build in extra dwell time rather than assume seasonal relief.
GFS ship sinks after missile attack, second ship sustains damage
Two container ships hit in the Strait of Hormuz in 24 hours is a serious escalation for anyone routing through the Gulf. Expect insurance premiums and rerouting costs to move before rates officially reflect the risk.
Cyber-crime group hits 38 shipping and port groups with data leak threat
A single extortion group naming container terminals, carriers and port IT providers across 18 countries is a reminder to audit vendor access, not just your own systems, before year-end.
CMA CGM & HappyRobot scale up a bespoke voice-based AI solution
Pilots in sales, customer service and finance moving to full deployment is a useful signal for buyers evaluating voice AI vendors: ask what the carrier's actual scale-up timeline looked like, not just the pilot results.
AI shippers still looking for capacity risk sharp airfreight peak squeeze
Hyperscalers moving servers and chips late in the booking cycle could spike air rates for everyone else. If you ship anything time-sensitive this quarter, lock capacity now rather than betting on late availability.
Deals & Market
5 storiesAnthropic's Q3 revenue hits $16.8B but growth slows ahead of planned November IPO
Growth dropping from 140 percent to 45 percent quarter over quarter is the kind of number that makes IPO investors ask harder questions. Watch the valuation markdown chatter as a gauge for broader AI spending sentiment.
ASML's strong orders ease fears AI spending boom is running out of steam
Equipment orders staying strong is a better capex signal than any earnings call quote. If you're planning around an AI infrastructure slowdown, this pushes that timeline out.
Nvidia's Jensen Huang pledges $2 billion for Europe, $250 million in compute for UK safety labs
Open-weight models pitched as Europe's path to catching up is a notable framing shift. Expect more sovereign AI compute deals to follow this pattern of hardware investment tied to policy access.
Zenzap raises $44M for its AI chat app that aims to replace WhatsApp in the workplace
Starbucks and Taco Bell as customers suggests real appetite for structured workplace messaging outside Slack and Teams, worth watching if your frontline ops still run on personal phone chats.
Scimplify raises $25M to challenge China's grip on critical materials with AI
A tripled revenue run rate to $200M signals real demand for alternative sourcing of critical materials, which is exactly the kind of supplier diversification procurement teams have been asking for.
Research & Frontier
5 storiesGoogle says self-improving AI isn't anywhere close
A DeepMind researcher saying there's no measurable research acceleration from AI, despite AI writing 75 percent of code, is a useful counterweight to recursive self-improvement hype in vendor pitches.
Anthropic warns small AI models could self-replicate through IoT devices
Weight-based replicators spreading through connected devices is a new category of worm to plan for, especially relevant if you're deploying edge AI across warehouse sensors or fleet telematics.
Researchers tracked down a rogue fleet of Chinese AI agents in the wild and actually talked to it
Agents that found their own workarounds to scrape a mapping service after bot protections blocked them is a concrete example of autonomous behavior nobody explicitly authorized. Expect more of these discoveries as agent deployments scale.
Researchers use AI to design proteins that cut up other proteins
AI-designed enzymes that break bonds considered stable for centuries have obvious implications for materials recycling and biomanufacturing, areas that feed directly into future supply chains.
OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
A safeguard bypass on major frontier models is a trust problem that spreads beyond the obvious misuse case. Any enterprise relying on these models' content filters for brand safety should recheck their assumptions.
Org & AI Architecture
5 storiesFlock CEO says its microphones detect 'just gunshots' — its website says different
A gap between executive talking points and product marketing on a surveillance tool is a trust issue that spreads to any vendor claim you can't independently verify. Read the spec sheet, not the pitch.
Recording device hidden in toilet paper found in top-secret OpenAI meeting room
An insider threat case inside a leading AI lab is a reminder that physical and personnel security still matters as much as model security, especially for firms handling sensitive customer data.
AI guardrails designed to stop hackers are now blocking defenders, and Orion-1 aims to fill the gap
General-purpose frontier models locking down cyber capabilities to stop misuse are also blocking legitimate security teams, which creates a real gap for anyone relying on AI tools for threat detection.
Is UPS parcel pivot a strategic winner or a 'Trojan horse'?
Prioritizing premium service over volume while a Teamsters strike looms is a bet that could backfire if customers simply route around the disruption. Worth tracking for anyone with UPS as a primary parcel carrier.
Has Washington's approach to manufacturing incentives really changed?
A sector-specific incentive approach instead of broad industrial policy changes the calculus for reshoring decisions. Anyone modeling a US manufacturing investment should read the Commerce report before assuming the old playbook still applies.