🗞️In this edition

• Google unveils Ironwood, its strongest AI chip

• Microsoft building a human-centered superintelligence

• Zuckerberg and Chan have a new AI mission

• Workflow Wednesday #44: AI for Decision-Makers

• In other AI news –

  • Ant Group bets on AI healthcare as China’s population ages

  • Perplexity to pay Snap $400M to power search in Snapchat

  • Meta launches AI-generated short video feed in Europe

  • 4 must-try AI tools

Google launches Ironwood TPU as Anthropic commits to 1 million chips

What's happening:

Google Cloud unveiled Ironwood, its seventh-generation Tensor Processing Unit, alongside expanded Arm-based computing options. The chip will become generally available in coming weeks.

Anthropic disclosed plans to access up to one million of these TPU chips, a commitment worth tens of billions of dollars and among the largest known AI infrastructure deals.

Ironwood delivers more than 4x better performance for training and inference compared to its predecessor. A single Ironwood "pod" connects up to 9,216 chips through Google's Inter-Chip Interconnect network operating at 9.6 terabits per second.

The 9,216 chips share access to 1.77 petabytes of High Bandwidth Memory. "Ironwood Pods can deliver 118x more FP8 ExaFLOPS versus the next closest competitor," Google stated.

Optical Circuit Switching technology automatically reroutes data traffic around failures within milliseconds. Google reported fleet-wide uptime for liquid-cooled systems has maintained approximately 99.999% availability since 2020, equivalent to less than six minutes of downtime per year.

Anthropic will have access to "well over a gigawatt of capacity coming online in 2026," enough electricity to power a small city. The company cited TPUs' "price-performance and efficiency" as key factors.

Google framed announcements around "the age of inference," where companies shift from training frontier models to deploying them in production serving millions of requests daily.

Why this is important:

Anthropic committing to one million TPU chips validates Google's custom silicon strategy against Nvidia's 80-95% market share dominance.

The deal represents a multi-year contract worth tens of billions. That's among the largest cloud infrastructure commitments in history and signals major AI lab confidence in non-Nvidia hardware.

Google's "age of inference" framing is strategic positioning. Training gets headlines, but inference is where sustained revenue happens. Every ChatGPT query, every Claude conversation, every AI-generated image is inference workload.

4x performance improvement over previous generation plus 118x advantage over "next closest competitor" are massive claims. If accurate, Google's closing the gap with or potentially surpassing Nvidia on inference-specific workloads.

99.999% uptime is table stakes for production AI. Less than six minutes downtime yearly means enterprises can depend on TPUs for mission-critical applications.

The gigawatt of capacity Anthropic's accessing shows infrastructure scale. Modern AI isn't measured in servers or chips anymore. It's measured in power consumption rivaling small cities.

Our personal take on it at OpenTools:

This is Google's strongest challenge yet to Nvidia's AI infrastructure monopoly.

Anthropic betting tens of billions on TPUs instead of Nvidia GPUs validates that custom silicon can compete on performance and economics. That's existential threat to Nvidia if other major AI labs follow.

The one million chip commitment is staggering. For context, most large AI clusters run 10,000-50,000 accelerators. Anthropic's accessing 20-100x that scale. That's nation-state level infrastructure.

The 4x performance jump generation-over-generation is impressive if it holds in production. But "118x more FP8 ExaFLOPS versus next closest competitor" needs context. What competitor? What workload? Marketing claims need independent validation.

99.999% uptime is genuinely impressive at gigawatt scale. That's operational excellence, not just good hardware. Google's decades running global infrastructure shows.

The real test is whether other frontier labs follow Anthropic's lead or this remains one major partnership. If OpenAI, Meta, or Mistral announce similar TPU commitments, Google wins. If it's just Anthropic, it's one big customer, not a market shift.

Read More

Microsoft forms team to develop humanist superintelligence

What's happening:

Microsoft AI wants you to know its work toward superintelligence involves keeping humans "at the top of the food chain." In a lengthy blog post Thursday, Microsoft AI head Mustafa Suleyman highlighted the formation of a new team dedicated to developing "humanist superintelligence" that's "designed only to serve humanity."

This superintelligence, according to Suleyman, won't be "an unbounded and unlimited entity with high degrees of autonomy" and will instead be "carefully calibrated, contextualized, within limits." 

Though Suleyman's blog says Microsoft AI will "reject narratives about a race to AGI," competition between Microsoft and OpenAI is about to get more intense. 

Under a new deal with OpenAI, Microsoft can now "independently pursue AGI alone or in partnership with third parties." Microsoft is "perfectly within its legal rights to use OpenAI's IP to develop its own AGI and attempt to win the race."

Suleyman has vision for "humanist" superintelligence with three main applications: serving as AI companion helping people "learn, act, be productive, and feel supported," offering healthcare assistance, and creating "new scientific breakthroughs" in clean energy.

"At Microsoft AI, we believe humans matter more than AI," Suleyman writes. "Humanist superintelligence keeps us humans at the centre of the picture. It's AI that's on humanity's team, a subordinate, controllable AI, one that won't, that can't open a Pandora's Box."

Why this is important:

Microsoft announcing a team dedicated to "humanist superintelligence" the same week it gains independence from OpenAI isn't a coincidence.

The new OpenAI deal lets Microsoft pursue AGI independently using OpenAI's IP. That makes Microsoft and OpenAI direct competitors for the first time. 

Suleyman's framing Microsoft's approach as philosophically different before that competition intensifies.

The "subordinate, controllable AI" framing is safety positioning. As Microsoft builds more powerful systems, they're preemptively addressing concerns about uncontrollable superintelligence.

Three applications, AI companion, healthcare, and clean energy, are broad enough to justify almost any product Microsoft wants to build.

Our personal take on it at OpenTools:

This is a corporate AI safety theater.

"Humanist superintelligence" that's "subordinate and controllable" sounds reassuring. But it's marketing language, not a technical specification. How do you ensure superintelligence remains subordinate? What makes it controllable? Suleyman doesn't say.

"Reject narratives about a race to AGI" is laughable when Microsoft just secured rights to independently pursue AGI using OpenAI's IP. That's joining the race, not rejecting it.

"Won't, can't open a Pandora's Box" is a confident assertion about technology that doesn't exist yet. Nobody knows how to build controllable superintelligence. Claiming Microsoft's figured it out before building it is premature.

Suleyman's smart to position Microsoft's approach as safety-focused before capabilities race intensifies. But words are cheap. The test is whether Microsoft actually builds different systems or just describes them differently.

Read More

Zuckerberg redirects philanthropy from social causes to AI disease research

What's happening:

Mark Zuckerberg and Priscilla Chan are shifting most of their philanthropy into curing or preventing diseases by capitalizing on AI advances, deepening the couple's shift away from social causes they once championed.

Zuckerberg and Chan said Thursday most charity would be invested into Biohub, a network of research labs in San Francisco, New York, and Chicago that aim to blend insights from biology and AI to advance healthcare. The couple acquired the team at AI startup EvolutionaryScale and tapped co-founder Alex Rives as head of science.

"Accelerating science is the most positive impact we think we can make. So we're going all in on AI-powered biology for our next chapter," Chan and Zuckerberg said. "Going forward, Biohub will be our primary philanthropic effort and where we'll dedicate the vast majority of our resources."

The group launched the Virtual Immune System project, which aims to model the human immune system to help simulate immune therapies, reprogram dysfunctional cells, and prevent diseases before they arise.

Why this is important:

This is billionaire philanthropy responding to political pressure by abandoning controversial causes.

Shifting to AI-powered disease research is politically safe. Nobody opposes curing diseases. Republicans and Democrats both support biomedical research. That's the point.

The 2020 "Zuckerbucks" controversy over $400M election administration funding clearly shaped this pivot. Republicans' sustained criticism made Zuckerberg retreat from civic engagement entirely.

Our personal take on it at OpenTools:

This is what happens when billionaire philanthropy meets political backlash.

Zuckerberg started with ambitious social programs: criminal justice reform, immigration, affordable housing, education for disadvantaged kids. All messy, political, hard problems requiring long-term commitment.

Then Republicans accused him of bias over election funding. Black employees said CZI marginalized their voices. Social programs became a political liability.

So he's pivoting to AI-powered disease research. Safe, bipartisan, generates positive press. Nobody criticizes you for trying to cure cancer.

"Going all in on AI-powered biology" sounds visionary. It's also abandoning communities that depended on CZI programs.

The Virtual Immune System project might cure diseases. That's valuable. But it's also easier than fixing criminal justice or housing policy. Science is less politically fraught than social change.

Zuckerberg's following Gates playbook: exit controversial domestic programs, focus on disease research with global impact. Malaria doesn't vote. Cancer doesn't have a political party.

The irony is Zuckerberg built Meta on connecting people and shaping social discourse. Now his philanthropy is retreating from social problems into lab-based research because social engagement got too messy.

Read More

This Week in Workflow Wednesday #44: AI for Decision-Makers – Workflows for Leaders & Execs

This week, I’ll show you how to use Perplexity to turn hours of competitor research into a 5-minute executive briefing. Think of it as a strategy analyst that never sleeps.

Workflow #1: Out-Research Your Competition with Perplexity (Perplexity Pro)
Step 1: Pick a competitor or market you want to monitor — like “Jasper AI” if you’re in the AI writing space.
Step 2: Ask Perplexity: “Give me a competitive……..We dive into this Perplexity workflow and two more AI-powered decision-making systems in this week’s Workflow Wednesday.

📢 What else is happening?

• Ant Group picks AI-assisted healthcare as new growth area amid China’s rapid ageing – The elevation of the business marks one of Ant’s most significant pivots since its cancelled IPO five years ago.

• Perplexity to pay Snap $400M to power search in Snapchat – The deal gives Perplexity exposure to more than 940 million Snapchat users, who will get answers from its AI engine when they ask questions to the company’s My AI chatbot.

• Meta brings its short-form video feed of AI slop to Europe – You can create and share short-form AI-generated videos and access a dedicated feed that displays AI videos from others.

👩🏼‍🚒Discover mind-blowing AI tools

1. Please Don't Code - A tool that aims to make coding for embedded systems, specifically Arduino, faster and easier

3. Publer - Allows users to manage, collaborate, and analyze their posts across multiple social media platforms

4. LOVO AI - An AI voiceover & text to-speech platform that offers a library of over 500+ human-like voices in 100 languages

5. Kittl - Helps users create stunning designs with AI-generated images and clipart