GPT-5.5 Goes to Work: Excel Integration, Ads, and a New Network Protocol

Spreadsheets Get a Brain

The strongest GPT-5.5 kernel is now embedded directly in spreadsheet software. The office suite integration can write complex formulas, explain data logic, and analyze tables at speed — the kind of work that normally bounces between a junior analyst and three Stack Overflow tabs.

Early reactions on social media suggest this is hitting a nerve. Spreadsheets remain the universal interface of business, and putting a capable model directly inside them is a distribution move that bypasses the chatbot tab entirely. Users don’t need to know they’re using AI — they just ask the spreadsheet to explain why Q3 revenue dipped and it does.

Ads Arrive

OpenAI launched its advertising platform, and the contrast with the Excel integration couldn’t be sharper. This is the monetization engine: models default to GPT-5.5 with noticeably faster response times, the entry threshold sits at $50,000, and traffic pricing runs roughly 3× Facebook rates.

The pitch is that memory-based targeting outperforms traditional search ads — the system knows what a user has been discussing, not just what they searched for five seconds ago. Whether users are comfortable with that trade-off is a different question. The industry appears to be bifurcating into traffic gateways (monetized through ads) and productivity tools (monetized through subscriptions), with ChatGPT trying to be both.

MRC: Making 10,000 GPUs Play Nice

On the infrastructure side, OpenAI, NVIDIA, and Microsoft jointly released the MRC Protocol — a network specification designed to solve the congestion that plagues large-scale training runs.

When you string together 10,000 GPUs, the bottleneck isn’t compute — it’s the network fabric between them. Packets collide, GPUs sit idle waiting for gradient updates, and expensive silicon runs well below capacity. MRC addresses this at the protocol level. Microsoft and Intel have already deployed it on their clusters with measurable improvements in training efficiency.

The Bigger Picture

Sam Altman separately hinted on X that ChatGPT “feels incredibly responsive now,” sparking speculation of a stealth model upgrade. Whether it’s the model, the infrastructure, or both — the combination of Excel distribution, ad monetization, and protocol-level infrastructure work suggests OpenAI is executing on all three layers of the stack simultaneously.