Microsoft starts merging Copilot apps into a super app
Three stories this afternoon: Microsoft begins the long-promised Copilot consolidation, Writer ships a cheaper flagship agentic model, and CNBC gets inside SK Hynix's $720 billion bet that AI memory demand is permanent.
Microsoft is finally combining its consumer Copilot and Microsoft 365 Copilot apps into a single interface — the first concrete step toward the "super app" it has promised for later this year. The unified app keeps the "Microsoft Copilot" name, gets a new icon, and will hold both personal and work accounts, with chats and content merged for people who used both apps, though data stays separate between account types. Mobile and web rollout begins mid-August, with Windows and Mac apps following in mid-September; enterprise security, compliance, and government controls are unchanged. Microsoft is retiring Copilot Podcasts and Deep Research after August 18 and won't carry Group Chat threads over — a reminder that consolidation has a cost. The move kills the two-Copilot-icons-in-the-taskbar annoyance, but the bigger signal is strategic: one surface for work and personal AI under a single product leader, laying the foundation for whatever "super app" arrives next.
Writer released Palmyra X6, a flagship agentic model for marketing and revenue teams, alongside agent-platform upgrades it says cut operating costs by 52% and improve speed by 48%. The model, post-trained on Z.ai's open-source GLM 5.2, scores 0.87 in Writer's own nine-evaluation suite at $2 per million input tokens and $8 per million output tokens — edging Claude Opus 4.8 (0.86) and GPT-5.5 (0.80), per the company's numbers. It can work unattended for up to eight hours, and new Playbooks and Skills give administrators a view of what agents are doing and what they cost. The timing is pointed: we covered DeepSeek's own API price hike this morning — DeepSeek hikes API prices up to 4.7x with new peak-hour billing — while Writer's pitch is that agent economics, not raw capability, is the next battleground.
CNBC got the first on-camera tour of SK Hynix's Yongin Cluster, the centerpiece of a $720 billion bet that AI demand for memory is permanent. The first of four fabs — a six-cleanroom tower about the height of a 50-story building — goes into production in February 2027, part of a national plan to double South Korea's memory output within five years. SK Hynix held 58% of the high-bandwidth memory market in the first quarter, according to Counterpoint Research, raised a record $26.5 billion on the Nasdaq in July, and has signed 10 long-term supply agreements including a $500 billion partnership with Nvidia. Chairman Chey Tae-won's framing — "It's like a war" — captures how far the boom-and-bust memory business has moved toward guaranteed demand; we flagged the squeeze before — AI demand sold out all 2027 DRAM and HBM capacity — and this buildout is the industry's answer: bet bigger, build faster, and hope the AI wave outlasts the cycle.
What to watch: Microsoft's super app launches later this year — watch whether the unified Copilot becomes the default AI entry point on Windows.
Does a single Copilot super app make Microsoft's AI easier to use — or just harder to escape? Tell us in the comments.
Sources: The Verge · GeekWire · SiliconANGLE · Writer · CNBC · SK hynix
