Microsoft Research outlined Project Silica, which is an effort to use glass storage to store data for up to 10,000 years. Microsoft said: "Glass storage is still emerging (experts predict it will require 3-4 more developmental stages before it’s ready for commercial use), but its benefits are clear: it’s durable, sustainable, and cost-effective." The project was published in Nature.

I'll just leave this link here. In a nutshell, here's your new org chart with your LLM overlords. We hope this is a joke, but don't think it is.

Etsy has sold Depop, a gen-Z thrifting marketplace, to eBay in a deal valued at $1.2 billion. Etsy acquired Depop in 2021 for $1.6 billion.

Etsy said it will focus on its core marketplace. eBay said it will take Depop to improve its position in consumer to consumer fashion commerce. Depop had 7 million active buyers with 90% of them under age 34 and 3 million active sellers.

Figma reported better-than-expected earnings and something rare occurred--the stock went up. Figma shares saw a 18% after hours bounce to recoup some of the 35% year-to-date decline.

Figma reported a fourth quarter net loss of $226.6 million, or 44 cents a share. Non-GAAP earnings were 8 cents a share, two pennies a share better than estimates. Revenue in the fourth quarter was $303.8 million, up 40% from a year ago.

For the first quarter, Figma projected revenue between $315 million to $317 million, up about 38% from a year ago. The company said 2026 revenue will be between $1.366 billion to $1.374 billion, up about 30%.

CEO Dylan Field said:

"Our growth and momentum show that our strategy is working. As AI gets better, Figma gets better—and we’re shipping faster than ever. In 2025, we expanded from 4 to 8 products and launched over 200 features, including new AI-native functionality."

The Royal Society Michael Faraday Prize Lecture was delivered by 2025 winner Professor Michael John Wooldridge. The talk is worth a listen. Key themes:

• The paradox of modern AI. Modern AI systems appear as articulate, confident experts, yet they frequently fail basic tests of rational intelligence.
• Risk of a "Hindenburg-style disaster." Wooldridge argues that intense commercial pressure is forcing tech companies to release tools before their flaws are understood. He suggests that a high-profile failure—such as a deadly self-driving car update or an AI-powered financial collapse—could shatter global trust in technology, much like the Hindenburg disaster ended the era of airships.
• LLMs struggle with simple tasks. AI can be incredibly effective at complex tasks but fail catastrophically at simple ones. He describes these systems as "glorified spreadsheets" rather than conscious entities.
• Anthropomorphism is dangerous. Pursuing humanlike AI is a "dangerous path" that misleads users into trusting them as reliable peers.
• What's next? Researchers are creating a field of experimental AI just to understand the unpredictable nature of systems already built.

The list of technology giants forming partnerships, building infrastructure and developing AI ecosystems is pretty long at this point. Why? The India AI Summit is underway. Here's an aggregation of developments via Techmeme to get the headlines at a glance.

Google I/O will be held May 19 and May 20 at the in Mountain View, CA. The event will feature a new Android as well as updates on XR and smart glasses. Google has a partnership with Warby Parker so it's likely we'll see some product launches. You can also expect Gemini updates. I'll be at Google Cloud Next, which kicks off a month before.

Mistral AI will acquire cloud provider Koyeb in a move to accelerate its Mistral Compute service. In a blog post, Koyeb said:

“The Koyeb platform will continue operating. In the coming months, we’re expecting to transition to become a core component of Mistral Compute and will double down on our Inference, Sandboxes and serverless capabilities for MCP Servers.

As we join Mistral AI and prepare for this transition, we're going to focus on customers on the Pro plan and above.”

Nicolas Bustamante, CEO of Fintool, penned an article on X on the AI-driven (actually more like Anthropic) software sell off and included a grid of what categories of SaaS will take the biggest hit. The article is worth a read to gauge the nuance of what SaaS vendors will be disrupted. Bustamonte riffs on the 10 moats of vertical software and what LLMs will do to each.

More:

• How three hotel giants are playing agentic commerce, AI
• How CEOs are answering the dreaded LLM disruption question
• SaaS under the microscope: How to evaluate your vendor
• SaaS death knell storyline illogical, but margin compression is here
• Observability stands out in enterprise software wreckage

Google Cloud and Unilever announced a five-year partnership that will revolve around bringing AI to consumer product goods. Unilever will use Google Cloud's enterprise AI stack for brand discovery, measurement and marketing. Unilever will build an AI-first digital backbone and develop agentic workflows using Google Cloud.

Nvidia is starting to talk about inference efficiency and total cost of ownership citing reports on low-latency agentic coding. It's highly likely you'll hear a lot more about inference total cost of ownership talk from Nvidia leading up to GTC in March.

Manus launched Manus Agents, a way to use Manus agents inside of messaging apps. Telegram is the first supported channel and it's safe to say WhatsApp will be a fast follow. After all, Meta now owns Manus.