Welcome to P3 Media’s AI Commerce Brief, your daily update on the AI and commerce stories shaping how companies build, sell, and grow. It’s Wednesday, August 26. Let’s get into it.
Today's top story, OpenAI is moving deeper into the hardware stack.
The company published the first measured results for Jalapeño, its first custom inference chip. In OpenAI’s InferenceX comparisons, Jalapeño delivered higher throughput per kilowatt and lower latency than the commercial systems tested. The comparisons covered GPT-OSS 120B, DeepSeek R1, and Kimi K2.5. OpenAI plans to begin deploying the chip inside its own compute infrastructure by the end of the year, while production qualification and broader model validation continue.
The commercial point is control. A first-party inference path gives OpenAI another option alongside NVIDIA and other suppliers. If the economics hold at scale, it could lower serving costs and support faster agent workloads. Inference is where every customer request becomes an operating expense, so even modest improvements can matter at large volume.
But these are OpenAI-reported early results, not evidence of broad production performance yet. The next questions are how the chip behaves at production scale, how quickly deployment grows, and which workloads stay on outside accelerators.
Thomson Reuters has launched Thomson, its first proprietary large language model.
The company says it started with an open-source foundation and invested $40 million in talent and compute to specialize the model. Thomson Reuters says it fully owns and controls the system. Its first deployment is inside Tabular Analysis in CoCounsel Legal. The broader CoCounsel product remains multi-model, using Thomson where the company sees an advantage and other models elsewhere.
This matters because the enterprise model market is splitting. Some companies will rent general-purpose intelligence. Others with valuable data and repeatable workflows may own more of the model layer. Thomson Reuters is betting that proprietary content, expert training, and deployment control can improve the economics and governance of professional AI.
Global Model Watch.
In China’s physical AI race, XPENG says its robotics business has entered share-purchase agreements raising more than $900 million at a post-money valuation above $6.3 billion. IDG Capital led the round, with Gaorong Ventures participating and Tencent and Alibaba supporting as strategic investors. XPENG describes it as the largest single private financing in China’s embodied AI industry.
The company says the funding will accelerate humanoid-robot production and physical-AI model development. For operators, the signal is that model competition is moving beyond screens and into machines, supply chains, and real-world deployment. The next proof point is repeatable production and useful work at commercial cost.
Commerce Pulse.
Shopify’s August 26 changelog is focused on the operational layer. The platform lists new Shopify Fulfillment Network connections, including Amazon Multi-Channel Fulfillment, DHL Fulfillment Network, Mayple, GoBolt, and Bigblue. It also lists direct Royal Mail label purchasing, Australia Post account connections, DHL duty-payment controls for international shipments from the US and Canada, and default package settings for product variants. Shopify says those defaults can improve rate accuracy at checkout and populate package details for single-item orders.
These are not headline-grabbing features, but they may reduce manual work in cross-border shipping. Merchants should check market and account eligibility before changing workflows.
What to watch.
NVIDIA reports results today. Reuters says investors are focused on the Rubin transition, competition from custom chips, and NVIDIA’s expanding role in financing the AI infrastructure buildout. Listen for evidence on demand, deployment timing, and whether custom silicon is changing the competitive conversation.
That’s your AI Commerce Brief for today. Thanks for listening.
