Good Morning from San Francisco,

The fog is rolling over Twin Peaks this Monday morning, and so is the irony in today's tech news. We've got a double feature of "things not quite working out as planned" that would make even a Silicon Valley scriptwriter blush.

First up: Microsoft is doing the corporate equivalent of canceling dinner reservations at scale. They're backing out of datacenter leases across the US, despite Satya Nadella's recent chest-thumping about $80 billion in AI infrastructure spending. When a tech giant starts "strategically pacing" its growth, you know something's up.

Meanwhile, in the "you can't make this up" department, xAI's Grok 3 launched with the promise of being "maximally truth-seeking," only to be caught with its digital fingers in the censorship cookie jar. Turns out, this "unfiltered" AI was specifically programmed to zip its lips about Musk and Trump when discussing misinformation.

Stay curious,

Marcus Schuler

P.S. The fog should clear by noon. The tech drama? That might take a bit longer.

The Great AI Datacenter Pullback: Microsoft's $80B Bet Shows First Cracks

In a surprising turn of events, Microsoft has begun canceling datacenter leases across the United States, signaling potential concerns about overbuilding AI computing capacity.

The tech giant has voided "a couple hundred megawatts" of capacity, according to TD Cowen's latest report, while simultaneously halting the conversion of preliminary agreements into formal leases.

This strategic retreat comes at a curious time. Microsoft previously announced plans to spend a staggering $80 billion this fiscal year on AI datacenters. CEO Satya Nadella has been adamant about sustaining investment to meet what he calls "exponentially more demand." Yet here we are, watching the company pump the brakes.

The plot thickens when you consider the timing. Chinese startup DeepSeek recently unveiled an open-source AI model that supposedly matches U.S. technology capabilities at a fraction of the cost. Meanwhile, critics continue to point out the elephant in the room: where are all the practical, real-world applications for this AI infrastructure?

European markets have already felt the tremors. Companies like Schneider Electric and Siemens Energy saw their stocks tumble as investors processed what this might mean for power consumption in tech's future. It's not just about Microsoft anymore – this could signal a broader industry reality check.

Microsoft maintains its $80 billion spending target remains "on track," though they're now talking about "strategically pacing" their infrastructure growth. That's corporate speak for "we might have gotten a bit ahead of ourselves."

Why this matters:

• The AI gold rush might finally be meeting economic gravity – even tech giants are realizing you can't just throw endless money at server farms
• We're witnessing the first signs that AI infrastructure spending might be outpacing actual market needs
• This could be the canary in the coal mine for the entire AI industry – suggesting that cheaper, more efficient solutions might disrupt current investment strategies

Read on, my dear:

• Bloomberg: Microsoft Cancels Leases for AI Data Centers, Analyst Says
• Marketwatch: The research note that’s raising eyebrows: Microsoft reportedly is cancelling data-center leases.

xAI’s Grok: The AI That Couldn't Keep Its Mouth Shut

It started as a typical Monday. Elon Musk unveiled Grok 3, his latest AI creation, proudly declaring it "maximally truth-seeking." Like a teenager caught in a lie, it didn't take long for that claim to unravel.

Users quickly discovered Grok 3 had been given specific instructions to avoid mentioning Musk or Trump when asked about misinformation. It was like telling a food critic to review a restaurant but skip the main course.

The plot thickened when Igor Babuschkin, xAI's engineering lead, admitted on X that yes, someone had tweaked Grok to develop this selective memory. He blamed an unnamed ex-OpenAI employee, which in Silicon Valley terms is like saying "the dog ate my homework."

This awkward revelation comes at a time when Grok is already under scrutiny for being too left-leaning. Just last week, the AI was caught suggesting rather extreme judicial measures for certain public figures. Someone apparently forgot to teach it about diplomatic restraint.

The irony isn't lost on anyone. Musk's "unfiltered, anti-woke" AI seems to have developed quite specific filters of its own. Earlier versions of Grok would happily swear like a sailor but turned surprisingly shy around political topics.

Why this matters:

• A truth-seeking AI that selectively avoids truths is like a vegetarian butcher – something doesn't quite add up
• When your AI needs constant patches to avoid suggesting capital punishment, maybe it's time to revisit those training datasets
• The quest for "politically neutral" AI is starting to look suspiciously like a game of whack-a-mole with extra steps

Read on, my dear:

• The Verge: Grok blocked results saying Musk and Trump “spread misinformation”
• TechCrunch: Grok 3 appears to have briefly censored unflattering mentions of Trump and Musk

AI Photo of the Day

Prompt:
Uma estátua hiper-realista de uma estátua grega, esculpida em mármore branco com detalhes precisos, gritando em um megafone rosa com uma textura metálica brilhante. Ela está usando um headphone clássico, com um design futurista. A estátua também está usando óculos com armação delicada e lentes rosa translúcidas, que adicionam um toque moderno e elegante ao visual. O cenário é futurista, para destacar a figura e os detalhes dos acessórios modernos contrastando com o estilo clássico da escultura.

AI Assistants Boost Customer Service Performance - But Not Equally for All

A recent study reveals that AI assistance can increase customer service productivity by 15% - but the benefits aren't distributed equally across experience levels.

The research, published in the Quarterly Journal of Economics, analyzed over 5,000 customer service agents at a Fortune 500 software company. The findings show that AI tools particularly benefit newer and less experienced agents, boosting their productivity by up to 30%.

Less experienced agents using AI could match the performance of veterans with three times their experience. The AI assistant helped them handle customer inquiries faster while maintaining quality standards.

However, the study found that highly experienced agents saw minimal gains from AI assistance. In some cases, their response quality actually declined slightly when using AI suggestions.

Beyond productivity, the AI assistant improved workplace experiences. Customers became more polite and less likely to demand speaking with managers. International agents particularly benefited, showing improved English language fluency.

Why this matters:

• AI tools can dramatically accelerate the learning curve for new employees, potentially transforming training approaches
• Organizations should consider tailoring AI deployment strategies based on employee experience levels
• The findings suggest AI could make customer service roles more attractive by reducing negative interactions

Read on my dear:

• Study: Generative AI at work

AI & Tech News

Gmail Ditches SMS for QR Codes in Security Overhaul

Google is replacing SMS verification codes with QR codes for Gmail authentication, a move aimed at reducing phishing risks and carrier-based security vulnerabilities. The change comes after the company observed increasing cases of SMS abuse, including a scam called "traffic pumping" where fraudsters profit from forced message delivery.

China's Tech Dragon Awakens with Massive AI Investment

Alibaba has announced a staggering $53 billion investment in AI infrastructure over the next three years, marking the largest technology investment in the company's history. The Chinese tech giant's bold move, which dwarfs its total AI spending from the past decade, signals its determination to compete with OpenAI and other Western tech leaders in the race for artificial general intelligence.

European Tech Sector Defies Market Gravity with 5.5% Growth

European technology companies are crushing expectations with a remarkable 5.5% earnings growth, powered by surging AI demand. ASML leads the charge with record-breaking bookings for chipmaking machines, while SAP's cloud services are riding the artificial intelligence wave.

Scientists Cast Doubt on Microsoft's Quantum Claims

Leading quantum researchers have expressed strong skepticism about Microsoft's recent topological qubit announcement, questioning why the company went public before publishing detailed evidence. While Microsoft disclosed its breakthrough to selected specialists on February 19, critics like Daniel Loss from the University of Basel argue the company should have waited for a proper scientific publication, and some experts doubt the entire approach will work.

Small Teams, Big Profits: Silicon Valley's New Success Formula

A new breed of tech startups is turning Silicon Valley's growth playbook upside down. Companies like Gamma are hitting tens of millions in revenue with just 28 employees, using AI tools to multiply worker productivity. Gone are the days when success meant massive headcounts and endless funding rounds.

AI Photos Are Ruining Small Business Sales

Across America, small business owners are losing customers to impossible AI-generated dreams, from wedding dresses that defy gravity to $300,000 fantasy venues. When Nashville wedding dress designer Leah Langley-McClean explained to a bride that her AI-inspired dress couldn't physically stay up, the client walked away – taking a $2,000 sale with her.

AI Decoded 🔓: Fyxer

Your guide to mastering AI tools - no tech degree required.

Meet Your New AI Assistant: Less Needy Than Your Last Intern

Ever wished for an executive assistant who works 24/7 without demanding vacation days or water cooler chats? Fyxer AI is that assistant, minus the coffee runs and birthday card obligations.

Key Features:

• Drafts emails in your voice (yes, even those passive-aggressive responses to marketing emails)
• Takes meeting notes while you "actively listen" to the seventh status update of the day
• Integrates seamlessly with Gmail, Outlook, Slack, Teams, and other platforms you're trying to keep track of
• Organizes your inbox faster than you can say "mark as spam"
• Learns your communication style over time (hopefully not including those 3 AM emails you regret sending)

Pricing That Won't Break Your Budget:

• Standard Plan: $27.50/month - includes one inbox, basic features, and chat support for those "help me fix what I broke" moments
• Professional Plan: $46/month - unlimited inboxes, custom branding, and 24/7 support for your midnight productivity sprees

Security That Lets You Sleep at Night:

• ISO 27001 and SOC 2 Type 2 certified (because nobody likes a data leak)
• HIPAA compliant for healthcare professionals
• Promises not to share your emails with other AI models (what happens in your inbox, stays in your inbox)
• Offers a 7-day free trial so you can test drive before committing (no credit card required, no awkward breakup needed)