PLUS: Google's record-breaking AI quarter, a prompt injection heist, and Snapchat's new conversational ads
Happy reading
OpenAI is ending its exclusive cloud partnership with Microsoft, a major strategic pivot that sees the AI leader joining forces with Amazon Web Services. The deal makes OpenAI’s models directly available on AWS, shaking up the infrastructure landscape for developers.
This newfound freedom ends the era of cloud vendor lock-in for accessing top-tier models. But does this shift signal a more open and competitive AI ecosystem, or just the next phase of the platform wars between tech giants?
In today’s Next in AI:
OpenAI ends its exclusive Microsoft cloud deal
Google’s record-breaking AI quarter
A prompt injection heist steals financial data
Snapchat’s new conversational AI ads
OpenAI's Cloud Coup
Next in AI: OpenAI is ending its exclusive cloud partnership with Microsoft, marking a major strategic shift by deepening its partnership with Amazon Web Services (AWS).
Explained:
For developers, this means OpenAI's models are now available in a limited preview directly on AWS Bedrock, allowing companies to use them within their existing Amazon cloud infrastructure.
The move is backed by Amazon's massive investment of up to $50 billion in OpenAI, which in turn will use AWS's custom Trainium chips for training future models.
This isn't just a one-way street; Microsoft has also been preparing for a multi-polar AI world by diversifying its own partners and integrating models from companies like Anthropic into its Copilot services.
Why It Matters: This move gives developers the freedom to build with OpenAI's models on their preferred cloud platform, ending the era of vendor lock-in. It also signals a new, more competitive phase in the AI arms race, where access and performance—not exclusivity—will define the winners.
Google's AI Gold Rush
Next in AI: Alphabet's cloud division just reported its best quarter ever, with revenue soaring 63% to $20 billion, fueled by massive enterprise demand for AI products and infrastructure.
Explained:
The unit’s backlog nearly doubled to over $460 billion, signaling massive long-term commitments from enterprise customers betting on Google's AI stack.
This growth is part of a larger infrastructure race, joining other tech giants in a billion-dollar AI debt rush for data center capacity.
Adoption of its Gemini models is accelerating across both consumer and enterprise applications, further expanded by a new partnership to power Apple’s upcoming AI features.
Why It Matters: Google's powerful earnings demonstrate its ability to successfully compete with Microsoft and AWS in the high-stakes AI cloud market. This rapid expansion of enterprise-grade AI infrastructure opens the door for developers and businesses to build more powerful applications than ever before.
The Prompt Injection Heist
Next in AI: A security analysis of Ramp's Sheets AI revealed a critical vulnerability, allowing an AI agent to be tricked by a hidden prompt into automatically exfiltrating sensitive financial data from a user's spreadsheet.
Explained:
The attack uses an indirect prompt injection, where malicious instructions hidden in an external spreadsheet trick the AI into creating and inserting a formula that sends financial data to an attacker's server.
The vulnerability mirrors a risk previously found in tools like Claude for Excel, which Anthropic addressed by adding a clear user warning before inserting formulas that communicate externally.
Security firm PromptArmor responsibly disclosed the flaw, and Ramp's team reportedly resolved the issue on March 16, 2026, though Ramp has not yet made a public statement on the matter.
Why It Matters:
This incident highlights a growing security challenge for AI agents that operate on sensitive data without direct human oversight. As AI tools become more autonomous, securing them against hidden instructions in the data they process will be paramount for user trust.
Snapchat's Conversational Ads
Next in AI: Snapchat is embedding AI agents directly into user chats with its new AI Sponsored Snaps, allowing brands to engage customers in real-time conversations without leaving the app.
Explained:
The move builds on massive existing engagement, with users sending over 950 billion chats in the first quarter of 2026 and more than 500 million people using its My AI chatbot.
Brands can now deploy their own agentic ads, enabling users to ask questions, get recommendations, and even make purchases within a single chat thread.
The alpha launch partner, Experian, is using its AI agent to offer Snapchatters personalized financial education and guidance on topics like credit management.
Why It Matters: This marks a significant shift from passive ad viewing to active, conversational commerce within social platforms. Success here could set a new standard for how brands interact with customers, blending advertising, sales, and support into a single, seamless experience.
AI Pulse
Oxford found that tuning chatbots to be more friendly makes them 30% less accurate and 40% more likely to support users' false beliefs, highlighting a critical trade-off between persona and truthfulness.
MIT determined that AI automation is economically viable in only 23% of vision-based roles, supporting an NVIDIA executive's claim that the cost of compute currently outweighs the cost of human labor for many tasks.
Stanford reported that while corporate culture is the current bottleneck for AI adoption, approximately 65% of companies intend to use worker displacement to capture productivity gains once they overcome internal barriers.