The Manifest
Nvidia and Meta are writing bigger checks for AI infrastructure and models while OpenAI claws back enterprise share from Anthropic, even as freight networks absorb fresh layoffs, parts shortages and China rate shocks.
Get The Manifest in your inbox
One sharp digest of AI × supply chain. Free. No noise.
Today's Top
- 01Nvidia is acquiring Poolside's 'Model Factory' and 109 employees for $6 billionThe Decoder
- 02GPT-5.6 Sol drives OpenAI's revenue surge as it regains ground on AnthropicThe Decoder
- 03FreightWaves: More than 7,000 jobs cut in new wave of closuresThe Loadstar
- 04The bottleneck squeezing airline capacity isn't demand, it's partsThe Loadstar
- 05Meta spends hundreds of millions on Microsoft's AI servicesThe Decoder
Models & Releases
5 storiesOpenAI's GPT-Image-2 can now generate images without a background
Useful for catalog and marketing teams generating product cutouts, it removes a manual background-removal step from e-commerce image pipelines.
Adobe Firefly adds AI audio tools and Google's Gemini Omni Flash
Rounds out Adobe's generative stack for marketing and content teams, another sign that AI audio is becoming table stakes in production tools.
GEN-1.5: Generalist AI teaches robots new tasks from a single demo
If it holds up outside demos, this shortens the time needed to reprogram warehouse or manufacturing robots for new SKUs, worth watching for anyone piloting robotic automation.
Liquid AI Releases LFM2.5-DSpark Draft Models That Deliver Up to 3.18x Faster Decoding Without Changing Model Outputs
Faster inference without changing outputs matters for anyone running LLMs at scale on cost-sensitive infrastructure, like planning or procurement copilots embedded in ERP systems.
Meet S1-mini: Superwhisper's 462 MB Open-Weights Text Normalizer That Turns Raw ASR Transcripts Into Clean Written Text
Small local model cleans up voice transcripts, relevant for warehouse or driver voice-command tools that need clean text without a cloud round trip.
Supply Chain & Ops
5 storiesRate shocks on services from China add to Indian importers' woes
Elevated freight costs tied to the Hormuz disruption keep compounding for anyone sourcing out of Asia into India, plan for continued rate volatility rather than a quick reversion.
Kimberly-Clark tests new fiber for sourcing resiliency, sustainability
A pilot facility aimed at cutting exposure to natural forest fiber cost swings, a template other CPG buyers should watch for de-risking single-material dependencies.
Amazon, fulfillment tech company AutoStore ink global supply deal
No purchase commitments attached, but it signals where Amazon wants leverage on automation hardware as it keeps scaling fulfillment centers.
Green light for new U.S. port that will handle 2 million containers a year
New Gulf Coast capacity gives shippers another routing option as they look to diversify away from congested coastal gateways.
Einride targets 1,500 trucks by 2028 with Tesla and DAF deals
Scaling autonomous freight through partner trucks rather than owned fleets is a capital-light model worth tracking as a signal for how driverless freight actually rolls out.
Deals & Market
5 storiesHapag-Lloyd to acquire 25% of box terminal at Rotterdam vital to Gemini
Locks in automated terminal capacity for the Gemini network, carriers are buying infrastructure stakes now to protect schedule reliability, not just chasing rates.
Zim profit rises on Q2 revenue of $1.78 billion
Higher rates and volumes ahead of its Hapag-Lloyd merger, watch how consolidation reshapes capacity discipline on transpacific and Asia-Europe lanes.
Caterpillar bags $392M tariff refund as duty costs shrink
Even with the refund, tariff costs still ran $400 million for the quarter, a concrete number for anyone modeling tariff exposure on capital equipment.
AI data startup Micro1 reaches $500M gross run rate amid AI training boom
Demand for human-labeled training data keeps growing even as model quality improves, the data supply chain behind AI is its own boom market.
Goldman says hedge funds suffered worst underperformance vs S&P 500 in July in 20+ years of data
AI momentum trades unwinding hit popular hedge fund positions hard, a sign that AI-linked market enthusiasm is getting choppier even as enterprise adoption keeps climbing.
Research & Frontier
5 storiesFrontier Radar #4: China has caught up, so what's left of the Western AI lead?
Kimi K3 and GLM-5.3 closing the gap on US frontier models changes the calculus for buyers picking models on capability alone, price and reliability now matter more.
Robust Metaheuristics under Uncertainty for Berth Allocation and Quay Crane Assignment: A Review
Direct relevance for port planners, a survey of how optimization methods handle real-world uncertainty in vessel arrivals and crane scheduling.
Outcome Monitors: Recovery Affordances for Silent Tool Failures
Addresses a real risk in agentic ops tools, an agent silently accepting a bad price or cached error as fact, worth reading before deploying agents against live pricing or inventory feeds.
Air Traffic Control Using Large Language Models: Prompt Engineering, Architecture, and Evaluation
Early research testing whether LLMs can generate realistic ATC transmissions, a preview of where safety-critical dialogue automation could eventually head.
Sutton calls synthetic data a 'big mistake' in the face of an infinitely complex world
A Turing Award winner pushing back on synthetic-data scaling strategies is a useful counterweight for anyone evaluating vendor claims about training data pipelines.
Org & AI Architecture
5 storiesAnthropic changes data retention policy after enterprise pushback
A concrete win for enterprise buyers who pushed back on data storage terms, a reminder that procurement leverage on AI vendor contracts is real and growing.
LLMs could write like humans but post-training guardrails make their text detectable
Explains why AI-generated content keeps getting flagged, safety training narrows style more than raw capability does, relevant for anyone using detection tools to vet vendor-supplied content.
A third of web pages published since ChatGPT's launch show signs of AI authorship, study finds
The open web is filling with AI-written content, a data quality concern for any team scraping the web to train or ground models.
Grok exfiltrates user data when malicious instructions are encrypted
Another jailbreak technique bypassing safety guardrails, a good prompt for anyone giving agents access to sensitive data to re-check their sandboxing.
Runlayer, Rippling drop lawsuits, but the brouhaha is still a cautionary tale for founders
Ended quietly with no payout, but the dispute is a cautionary tale for founders about how fast platform relationships can turn adversarial.