The Manifest
OpenAI is closing in on a $30 billion raise at a $1.25 trillion valuation while major US banks start locking out unauthorized AI agents, and a new Gartner survey shows most enterprises still cannot show real returns on their AI spend even as freight and rail earnings climb into peak season.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01OpenAI working on new $30B raise at $1.25T valuation ahead of IPOTechCrunch AI
- 02Major US banks are locking out AI agents that log in with customer credentialsThe Decoder
- 03UPS automates 4 additional facilities ahead of peak shipping seasonFreightWaves
- 04Only 13 percent of companies see real returns from AI, Gartner survey findsThe Decoder
- 05DeepSeek reportedly raises $15B days after US restrictionsTechCrunch AI
Models & Releases
5 storiesOpenAI working on new $30B raise at $1.25T valuation ahead of IPO
Investor appetite remains enormous even though the valuation target got trimmed to close the round. Watch how this capital flows into pricing and compute costs for the AI tools your teams already use.
Google Cloud Launches Agent Compute Engine: A MicroVM per AI Agent
Isolated microVMs per agent address a real procurement concern: credential exposure when agents act on your systems. Worth tracking if ops teams are piloting autonomous agents in production.
Claude Duet app is Anthropic's bet on agents
An always-on agent that works while you're offline, wired into Gmail, Slack and Stripe. The real test for operators is whether it cuts busywork without adding approval fatigue.
Canva releases its first image generation model
Lower-cost image generation built into a tool marketing and packaging teams already use. Incremental, but it lowers the barrier for in-house creative production.
OpenAI makes its Jev competitor widely available as decision model competition heats up
Decision models built for structured judgment calls could eventually fit into planning and QA workflows where you need fast, consistent rulings rather than open-ended chat answers.
Supply Chain & Ops
5 storiesUPS automates 4 additional facilities ahead of peak shipping season
A concrete capacity move heading into peak, worth watching for knock-on effects on dock throughput and labor planning at competing parcel networks.
K+N claims market share gains as 'strong' air cargo peak gathers pace
Most of the air cargo growth is coming from market share grabs, not organic demand growth, so expect continued rate pressure on the carriers losing ground rather than a broad based peak.
Containership charter rates at a four-year high despite record orderbook
Rates are climbing even as a record orderbook looms for 2028, a reminder that near-term tightness and medium-term overcapacity can coexist. Plan charter timing accordingly.
US trailer orders soar 385% as Class 8 truck market sends mixed signals
Carriers are front-loading equipment purchases ahead of expected price hikes. Worth watching if you're negotiating fleet replacement timing or equipment contracts.
SONAR launches new routing guide and contract rate data, plus expanded SearchCarriers monitoring
New visibility into what shippers actually pay once freight falls past the first tender is useful leverage for anyone renegotiating broker and carrier contracts.
Deals & Market
4 storiesDeepSeek reportedly raises $15B days after US restrictions
A massive round landing just days after federal agencies were told to remove the technology shows Chinese AI capital isn't slowing down. Worth a look if DeepSeek models sit anywhere in your vendor stack.
Norfolk Southern profits rise on higher volume and revenue
Record revenue on higher volume and fuel surcharges suggests rail demand held up better than expected this quarter, a useful data point for anyone benchmarking modal cost shifts.
Union Pacific posts record third-quarter revenue and profits
Pricing and volume both up despite inflation and fuel costs. Rail capacity looks less like a bottleneck than it did a year ago.
Stord expands network with AMS Fulfillment purchase
Another fulfillment roll-up adding East and West Coast hub coverage. Relevant if you're evaluating 3PL options for omnichannel distribution.
Research & Frontier
5 storiesLeft alone with nothing to do, AI agents started playing
A quirky MIT finding with a real point: agents given idle time got better at their actual tasks afterward. Worth considering before cramming every minute of agent uptime with work.
Benchmarking Model Compression Techniques Across Heterogeneous Edge Platforms: A Predictive Maintenance Use Case for Rotating Machinery
Direct relevance for plant and maintenance teams deciding which compressed models actually run well on the edge hardware they already have, rather than the hardware vendors wish they had.
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?
A sober benchmark on how far coding agents get on real scientific improvement tasks, useful context before trusting agents with R&D work beyond routine engineering.
AI and the Enclosure of the Scientific Commons
A debate worth watching if your organization relies on open research. The argument is that AI labs are quietly privatizing what used to be shared scientific infrastructure.
SpecEvo: Speculative Evolution with Large Language Models for Cost-Efficient Scientific Discovery
A method for cutting the cost of LLM driven experimentation loops, early stage but it points toward cheaper automated R&D pipelines down the line.
Org & AI Architecture
5 storiesOnly 13 percent of companies see real returns from AI, Gartner survey finds
Most CIOs still can't show ROI, yet budgets keep rising anyway. Leaders should be ready to explain that gap before the next budget cycle.
Uber puts all 30,000 employees on ChatGPT Work
A full company-wide rollout rather than a pilot. Worth watching for public reporting on what actually changes in day to day workflows once scale hits.
Major US banks are locking out AI agents that log in with customer credentials
Banks are discovering customers already hand their login credentials to AI agents without clear rules. Expect similar scrambles wherever your own systems accept credentials from an unverified actor.
Microsoft's Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases 'medical superintelligence'
The reported gains come mostly from the underlying models, not Microsoft's added system. A reminder to separate vendor marketing from what the base model already does.
Chinese hackers spied on thousands of companies after breaching IT giant ConnectWise
Attackers used AI agents to scale a single breach into thousands of downstream compromises. A concrete reason to tighten vendor access reviews now rather than after an incident.