Apple to Allow Rival AI Model Selection in iOS 27
According to Reuters, Apple has revealed plans to let users choose AI models other than Apple Intelligence in its upcoming OS, iOS 27. The change will span multiple features in iOS 27, with competing AI models such as ChatGPT and Gemini expected to be integrated at the system level.
Apple has until now positioned its own Apple Intelligence at the core of its AI features, but the decision to open up the platform appears to be driven by user demand and regulatory pressure. A wider choice of AI models is good news for users, but what it means for Apple's own AI strategy — and how things unfold from here — remains to be seen.
Coinbase Cuts 14% of Workforce, Citing AI Acceleration
According to CNBC, cryptocurrency exchange Coinbase has announced it will cut jobs amounting to 14% of its workforce. The company cited "AI-driven acceleration of work" as the reason, making this a case where AI adoption has translated directly into headcount optimization.
Large-scale layoffs attributed to AI are gradually increasing among tech companies. The fact that a prominent company like Coinbase explicitly named AI as a reason for the cuts suggests that AI's impact on employment is entering a concrete phase. That said, whether the eliminated work is actually being replaced by AI will require mid-to-long-term verification.
ServiceNow Unveils Autonomous AI Workforce Built with Microsoft and NVIDIA
According to Fortune, ServiceNow announced at its annual conference, Knowledge 2026, an Autonomous AI Workforce built in collaboration with Microsoft and NVIDIA. The offering is a set of AI agents capable of autonomously executing internal business tasks, positioning the company to significantly advance the automation of enterprise operations.
AI agent adoption in the enterprise space is accelerating rapidly, but it is noteworthy that ServiceNow used the phrase "an AI workforce that can run the company." However, details on the actual scope of autonomy and human oversight requirements have not yet been disclosed, so evaluation in real-world deployment remains to be seen.
OpenAI Moves to Develop an AI Smartphone, Mass Production Possibly in 2027
According to The Decoder, OpenAI plans to develop an AI smartphone as its first in-house hardware product. Based on information from analyst Ming-Chi Kuo, chips will be sourced from MediaTek and Qualcomm, with Luxshare expected to handle manufacturing. Mass production is slated to begin in the first half of 2027, with initial shipments projected at up to 30 million units.
Rather than the traditional app grid UI, the device is reportedly envisioned with an interface centered on AI agent task streams. That said, AI hardware has so far struggled to gain a foothold in the market, as seen with products like the Humane AI Pin and Rabbit R1. OpenAI's choice of the smartphone — an existing form factor — can be read as an acknowledgment that experimental form factors have yet to achieve widespread adoption.
OpenAI Makes ChatGPT Ads Self-Serve, Introduces CPC Bidding
In an official announcement, OpenAI expanded the ways advertisers can buy ChatGPT ads. The company introduced a beta self-serve Ads Manager, CPC (cost-per-click) bidding, and enhanced measurement tools, stating that it is focused on privacy protection and keeping ads separate from conversations.
As ChatGPT's monthly active user base continues to grow, advertising revenue is becoming an important monetization channel for OpenAI. Offering a self-serve platform lowers the barrier to entry for advertisers, and ChatGPT's presence in the advertising market looks set to grow even further.
Blackstone and KKR in Partnership Talks with Google on Enterprise AI Model Deployment
According to Bloomberg, major private equity firms Blackstone and KKR are in talks with Google over an AI model partnership. The move is seen as an effort to accelerate AI adoption across their portfolios by providing Google's AI models to portfolio companies.
Partnerships between private equity firms and AI providers suggest that enterprise AI adoption is moving beyond mere technology implementation into a stage of ecosystem formation tied to capital. With access to AI for portfolio companies potentially becoming a source of competitive advantage, similar partnerships could spread to other PE funds and AI companies.
SubQ: A Sub-Quadratic LLM Arrives for 12-Million-Token Contexts
SubQ, which gained traction on Hacker News, has announced an LLM with a sub-quadratic architecture capable of processing contexts of up to 12 million tokens. In conventional Transformer-based LLMs, computational cost grows quadratically as context length increases, but SubQ claims to keep this sub-quadratic, enabling ultra-long-context processing.
Twelve million tokens is roughly the amount of information contained in dozens of books. Long-context processing is one of the major challenges for LLMs, and if SubQ's approach works in practice, it could significantly change the need for RAG (retrieval-augmented generation). That said, independent verification of benchmark details and real-world inference performance is still limited at this point, so practical usefulness will have to await further evaluation.