Google Inside Meta

The Humbling of Meta

The story of the AI boom so far is that every tech giant is spending tens of billions of dollars to build its own world-beating intelligence. Mark Zuckerberg has been on a particularly aggressive campaign, poaching top talent for a secret "Superintelligence Lab" with the singular goal of winning this race. It is, then, a little weird that his company is now in talks to use the core technology of its fiercest advertising rival, Google.

This week, The Information reported that Meta staffers have had discussions about using Google's powerful Gemini AI models to improve the targeting capabilities of its own multi-billion-dollar ad business. Here is Reuters on the potential deal:

Meta staffers have had discussions with Alphabet's Google Cloud about the possibility of using its Gemini models to improve the Facebook parent's ad business, the Information reported on Thursday.

...The Meta employees have proposed fine-tuning rival Google's Gemini and open-source Gemma models on Meta's ad data to improve its ad targeting capabilities, the report said.

[Meta's] move to pick Google's AI over its own in-house models underscores the issues the company has had with scaling AI technology, despite the billions of dollars it has spent on research, infrastructure and talent.

This is a quiet but profound admission. The very fact that these talks are happening suggests that despite the massive spending spree, Meta's homegrown Llama models are not yet ready for the single most important job at the company: running its money-making machine.

The move reveals a critical distinction in the AI world. There is a massive difference between building a general-purpose AI that can write poetry and building a highly specialized, reliable system that can drive billions in ad revenue. While Meta's "Superintelligence Lab" is focused on the long-term, "moonshot" goal of building AGI, its core business needs the best possible technology today. The talks with Google suggest that, for now at least, the best technology for the job may not be its own.

The whole thing is a fascinating case of strategic pragmatism, and it is part of a much broader trend reshaping the tech landscape. The AI arms race, it turns out, is creating a world of strange bedfellows. Apple is relying on OpenAI to handle complex queries on the iPhone. OpenAI, in turn, is using infrastructure from Google and Oracle to reduce its dependence on Microsoft. And now Meta is considering using Google's AI to power its most critical business function. The sheer cost and complexity of building a best-in-class model for every single task is forcing even the biggest and richest companies to admit they can't do it all themselves. The result is a messy, interdependent ecosystem where your fiercest rival is also, sometimes, your most important technology partner.

The Quantum Internet

One way to bet on a gold rush is to pick the most promising prospector and hope they strike it rich. Another way is to sell the picks and shovels to everyone. This, in essence, is the strategy Cisco is now deploying for the strange and chaotic world of quantum computing.

While giants like Google, Microsoft, and IBM are pouring billions into a high-stakes race to build a functional quantum computer, Cisco has decided not to join the race at all. Instead, it's building the plumbing. This week, the company unveiled a new software tool designed to weave together all of these disparate, competing quantum machines into a single, functional cloud.

The basic situation is that the quantum world is a technological mess. There is no agreed-upon standard for how to build a quantum computer; IBM uses superconducting circuits, while others use trapped ions or photonics. They are all fundamentally incompatible. Cisco's new software acts as a universal translator, a smart layer that can take a complex problem, break it into pieces, and assign each piece to the specific type of quantum machine best suited to solve it. Here is Cisco's senior vice president Vijoy Pandey on the logic, via Reuters:

"You as the customer, as the quantum algorithm developer, should not worry about the kinds of technology that exist... We will handle that complexity."

This is, of course, the classic Cisco playbook. The company became a titan not by building better computers in the 1990s, but by building the routers and switches that allowed all the different, incompatible computers of that era to talk to each other. They built the internet. Now they are trying to build the quantum internet.

The bet is a strategic hedge on continued chaos. Cisco wins if the quantum hardware market remains a fractured landscape of competing technologies, a scenario where a universal connector becomes increasingly indispensable. It is a tacit acknowledgment that building the roads can sometimes be a much better business than betting on which car will win the race.

The Scoreboard

• AI: Nvidia is letting anyone use its AI voice animation tech (The Verge)
• AI: Judge in Anthropic copyright case preliminarily approves $1.5 billion settlement with authors (CNBC)
• Social Media: Trump approves TikTok deal through executive order (CNBC)