OpenAI Ships Reasoning Voice API While ASUS Loses Five Million Boards to AI Foundries

01OpenAI shipped a reasoning voice API with Parloa and Uber already running on it

OpenAI released new realtime voice models in its API and pushed out two enterprise case studies, Parloa and Uber, the same day. The bundle shifts how the company is selling voice: not a single model launch, but an agent stack with deployment references attached.

The new realtime models reason, translate, and transcribe inside one session, according to OpenAI's announcement. Developers building voice agents previously chained speech-to-text, an LLM, and text-to-speech as separate API calls, with latency stacking at every hop. The single-session design collapses that pipeline into one stream.

Reasoning inside the voice loop is the part developers will test first. A bot that can pause, work through a multi-step request, and respond without dropping the connection has been the gap between demo voice agents and production ones. Translation in the same session removes another orchestration layer for any product crossing languages. Transcription folded in means call analytics no longer require a separate provider.

Parloa builds voice agents for enterprise customer service. The company says its platform lets customers design, simulate, and deploy voice flows on top of OpenAI models, according to OpenAI's case write-up. Simulation is the part that matters. Voice agents fail in ways that are expensive to debug live, and tested rollout paths are what enterprise buyers ask for first.

Uber is using OpenAI on both sides of its marketplace. Drivers get an assistant aimed at helping them earn more, and riders get voice features for faster booking, according to the case study. The deployment runs against Uber's real-time global marketplace, which is the scale claim OpenAI is foregrounding.

Voice was already an OpenAI product line. What changed is the positioning. Two named customers running production workloads on the same architecture, announced the same day as the model, is the message OpenAI is sending to engineering teams.

For developers evaluating whether to integrate voice, the calculus moved. The build-vs-buy line used to sit at stitching three APIs together and tuning latency. It now sits at sending audio in, getting audio out, with reasoning held in the loop.

02OpenAI's president read his own 'greedy' diary entries to a jury, the same week ChatGPT started serving ads

Brad Lightcap, OpenAI's president, took the stand in Elon Musk's lawsuit against the company and was handed his own private journal. Musk's attorneys had him read passages aloud to the jury, then asked him to explain why they sounded, in the lawyers' framing, greedy. According to Ars Technica, Musk's legal team argued the entries captured the moment OpenAI walked away from its founding nonprofit mission. Lightcap defended the writing as personal reflection, not corporate strategy.

The same week, OpenAI announced it would start testing ads inside ChatGPT.

OpenAI says the ads will carry clear labels, will not influence model answers, will run with privacy protections, and will give users controls. Keeping the free tier free is the stated rationale. The blog post does not specify how ads will be selected, which advertisers appear in early tests, or what data ChatGPT conversations will share with the ad system.

The two announcements sit on the same timeline. Inside the courtroom, Musk's lawyers presented the journals as evidence of a commercial pivot OpenAI had publicly denied. Outside, ChatGPT was beginning to carry advertising, the type of move the lawsuit alleges was always part of the plan. Lightcap's response was that private journaling and corporate intent are not the same thing.

Musk filed the suit in 2024 alleging OpenAI breached its founding agreement by restructuring around a for-profit arm. The trial is the first time OpenAI executives have faced sustained cross-examination on the internal record. Private writings, board minutes, and email threads are now part of the court file. Musk co-founded OpenAI in 2015 and left the board in 2018.

The ad test is limited for now. OpenAI has not said which countries or user segments will see ads first, when paid tiers might see them, or when the test ends. Users can provide feedback during the trial, the company says.

03Five Million Motherboards ASUS Won't Ship in 2025. The Foundry Capacity Went to AI

Motherboard sales across the enthusiast PC market collapsed by more than 25% in 2025, according to a Tom's Hardware report describing the shortages as unprecedented. ASUS alone is projected to sell roughly 5 million fewer boards this year than its previous baseline. Gigabyte, MSI, and ASRock are all logging reduced shipments. The four names account for most of the DIY motherboard market.

Consumer demand is not the cause. Foundries fabricating the chipsets and discrete GPUs that feed self-built PCs have shifted wafer allocation toward AI accelerators, where order volumes and margins run higher, the report said. Tighter wafer slots upstream send fewer chipsets to board partners. Without chipsets, motherboards don't ship.

That capacity is heading somewhere visible. Frontier labs are now locking up entire data centers at a time. Anthropic disclosed at its Code w/ Claude keynote a deal with SpaceX/xAI to take "all of the capacity" of the Colossus facility, according to Simon Willison's notes on the announcement. Orders at that scale pull silicon out of the consumer channel before distributors ever see it.

The squeeze hits a specific buyer. Self-build enthusiasts and small system integrators have anchored the DIY motherboard category for two decades, depending on a steady flow of mid-range chipsets to keep configurations in stock. A 25% supply cut at the top of the funnel translates into thinner SKU shelves and longer lead times for builders. ASUS, the volume leader, absorbs the largest absolute hit at roughly 5 million units. The drag falls proportionally on Gigabyte, MSI, and ASRock across their lines.

Discrete GPU shortages have run alongside the motherboard contraction. Tom's Hardware described chipmakers as "strangling" enthusiast PC supply to build more AI chips. The four board makers are now competing for the wafer leftovers from a market they used to set the tempo in.