AI hardware innovation and global autonomous vehicle expansion continued as CoreWeave deployed NVIDIA Vera Rubin clusters and Waymo announced plans for Singapore robotaxis by 2028. Meanwhile, political friction intensified as Donald Trump rejected new AI regulations and clashed with top industry executives. At the same time, tech equities faced pressure after Larry Ellison scrapped a massive stock sale following market concerns over heavy AI infrastructure spending.
01Background
Trump Met Privately With OpenAI CEO Sam Altman at GOP Convention
The leader in frontier artificial intelligence is prioritizing safety-driven listing delays over immediate capital market access, contrasting with political resistance to federal industry guardrails.
President Donald Trump met privately with OpenAI chief executive Sam Altman backstage at the Republican midterm convention on Thursday. Altman requested the meeting to discuss frontier development and the growing power of the technology. The private discussion followed public warnings from former researchers regarding the safety risks of rapid artificial intelligence scaling. Altman stated over the weekend that OpenAI would delay going public this year due to safety concerns and argued that slowing development is worth the cost. Trump rejected calls for federal guardrails, warning on social media that industry regulations would lead to oblivion and while calling artificial intelligence the greatest economic development engine in history.
Waymo to Launch Robotaxi Service in Singapore by 2028
Exporting Alphabet's autonomous driver into tropical transit hubs tests whether its perception stack can scale internationally without incurring prohibitive, city-by-city re-engineering costs during manual mapping phases.
Alphabet-owned Waymo will launch an autonomous ride-hailing service in Singapore by 2028, marking its first entry into Southeast Asia. Waymo is partnering with the Ministry of Transport and Land Transport Authority to deploy an initial fleet of all-electric Jaguar I-PACE vehicles. The cars will arrive in the coming months, ahead of a readiness phase in 2027 where trained autonomous specialists will drive the vehicles manually to adapt the Waymo Driver to local road geometry and monsoon weather conditions. Fully autonomous rides available to the public through the app remain subject to regulatory approvals. The expansion follows similar international announcements for Tokyo in 2027 and London by the end of 2026, alongside domestic footprints spanning about 15 US cities with around 4,000 vehicles.
Trump Rejects Additional AI Regulations, Criticizes Anthropic CEO
Federal refusal to mandate AI safety standards shifts the regulatory burden entirely onto private balance sheets, favoring aggressive infrastructure spenders over safety-focused developers like Anthropic.
Donald Trump rejected additional regulations and criticized Anthropic CEO Dario Amodei. The stance aligns with pushback from other tech leaders opposing coordinated industry slowdowns. Market participants continue to evaluate the broader regulatory landscape for frontier models.
Deploying state-of-the-art models on consumer hardware collapses downstream inference costs for end users while preserving the massive capital expenditures required for frontier training infrastructure.
startup PrismML released Bonsai 2 27B, a compressed 27.8-billion-parameter language model built on Alibaba's Qwen3.8 27B architecture. The model is scaled down to 5.9 gigabytes using ternary weights, which restrict neural network values to -1, 0, or 1 to reduce memory usage by more than nine times compared to full-precision counterparts. Across a twenty- suite, the model achieved an aggregate score of 83.9, retaining 98.2 percent of the performance of the uncompressed base model. The release is available for free download under an Apache 2.0 license, running locally on consumer hardware like the NVIDIA GeForce RTX 5090 at speeds reaching up to 143 tokens per second. While the compression method reduces runtime memory requirements, PrismML trained the model on Google v5 TPUs, leaving heavy training infrastructure costs unchanged for frontier developers.
Bonsai Benchmark Score Retention (%)
Bonsai 2 27B closes the performance gap to 98 percent of the base model.
Anthropic Integrates Claude Chat and Cowork into Unified Interface
Automating tool selection across native document and slide creators positions Anthropic to capture enterprise seat value directly from Microsoft and Google rather than remaining a sidecar application.
Anthropic is combining its Claude chat and Cowork features into a single interface alongside the launch of new presentation and document tools. The update merges chat, Cowork, Artifacts, and the Claude Design tool into one window where users no longer need to switch tabs or choose a specialized feature for each task. The company is also introducing Claude Docs and Claude Slides in , allowing users to create documents and slide decks that can be exported to Google Docs, Microsoft Word, PowerPoint, or PDF files. The unified interface rolls out first to Pro and Max subscribers on web, desktop, and mobile, before expanding to Team and Free plans.
Larry Ellison Cancels Plan to Sell Up to $7.5 Billion of Oracle Stock
Halting an executive share-sale plan eliminates a major equity overhang while the market weighs expanding AI infrastructure capital expenditure against rising debt loads.
Larry Ellison canceled his plan to sell up to 50 million shares of Oracle stock worth about $7.5 billion. The executive chairman and co-founder adopted the trading plan on June 22 to run through October 24, but reversed the decision on Saturday, September 12. Oracle confirmed that no stock was sold under the plan and that Ellison has no other current plans to sell shares. The cancellation failed to lift the stock, with Oracle shares sliding about 4.7% to $143.24 on Monday. Investors instead focused on mounting costs tied to Oracle's aggressive buildout and broader pressures across the technology sector.
CoreWeave Deploys Multi-Rack NVIDIA Vera Rubin NVL72 Cluster
CoreWeave's deployment of multi-rack Rubin clusters demonstrates how specialized cloud providers use deep NVLink integration and localized storage acceleration to defend compute pricing against hyperscaler commodity infrastructure.
CoreWeave deployed a multi-rack NVIDIA Vera Rubin NVL72 cluster on CoreWeave Cloud, connecting hundreds of NVIDIA Rubin GPUs into a single scale-out cluster for agentic workloads. Each individual Vera Rubin NVL72 rack pairs 72 Rubin GPUs with 36 Vera CPUs, alongside NVIDIA NVLink 6, ConnectX-9 SuperNICs, and BlueField-4 DPUs. The company uses NVIDIA Spectrum-X Ethernet networking to unify multiple racks into one coordinated system. CoreWeave also added cross-region write acceleration and a lower-cost Archive tier to its AI Object Storage platform. Its Local Object Transport Accelerator delivers up to 7 gigabytes per second of throughput per while reducing read latency by eight times compared to traditional storage clusters. CoreWeave shares rose following the announcement.
Rapid hardware deployments and model compression are pushing local AI capabilities forward, even as heavy capital expenditure strains corporate valuations. Political pushback against safety guardrails leaves the future regulatory environment for next-generation frontier models entirely uncertain.
This, every morning.
SuMarket writes Technology Sector every morning, along with every other section of the market and the companies and topics you follow. Free to read.