Google claims Gemini 3 delivers PhD-level reasoning. The fine print admits 72% accuracy and minute-plus generation times. Early testing reveals graduate-student errors. OpenAI's GPT-5 disaster opened the door, but can Google's benchmarks justify $7T spending?
Good Morning from San Francisco, When the CEO warning about investment bubbles is spending $75 billion on the very thing
Google's CEO warns investors about AI bubble risks while Alphabet commits $75 billion to the same markets. The company ships error-prone systems requiring self-verification, delays climate targets despite record profits, and tells workers to adapt alone.
Good Morning from San Francisco,
When the CEO warning about investment bubbles is spending $75 billion on the very thing creating them, you're watching incentives override judgment in real time.
Today's newsletter examines two variations on that theme.
First, a market leader who sees the problem but can't stop making it worse.
Second, a competitor discovered the fastest path to the top of AI leaderboards involves systematically removing safety features other labs spent years building.
Both stories share one pattern: The market rewards exactly what industry leaders publicly identify as threats to long-term stability.
When competitive pressure meets stated principles, watch what people do, not what they say.
Stay curious,
Marcus Schuler
Sundar Pichai told the BBC this week that "irrational" investment patterns threaten AI markets, echoing Alan Greenspan's 1996 warning before the dotcom crash.
Alphabet simultaneously announced $75 billion in 2025 capital expenditures, 43% above analyst expectations, nearly all flowing into AI infrastructure that doubled the company's market cap to $3.5 trillion in seven months. The company posted $350 billion revenue and $100 billion profit for 2024 while cloud revenue missed quarterly targets, yet shares climbed 46% anyway on AI enthusiasm.
Pichai positioned Google's chip-to-consumer "full stack" control as insulation against turbulence while admitting "no company is going to be immune, including us."
The contradiction runs deeper: Google ships AI systems it acknowledges make errors, directing users to verify outputs with Google Search, which now surfaces AI-generated summaries requiring their own verification.
Alphabet also confirmed delaying its 2030 net-zero climate target as AI operations drove emissions up 48% since 2019. OpenAI's economics illustrate the bubble mechanics Pichai identified, $1.4 trillion in deals surrounding a company that lost $5 billion on $3.7 billion revenue in 2024, with leaked documents suggesting inference costs exceed income.
Why This Matters:
• Smaller AI companies lack Alphabet's hedged exposure to profit from both boom continuation and correction, concentrating risk asymmetrically across the sector.
• When market leaders delay climate commitments for growth, it establishes precedent allowing competitors to prioritize expansion over environmental goals through 2030.
Robert Brown
Prompt:
a duck wearing sunglasses and drinking boba tea, full-body shot, hip-hop style, white background, in the style of cartoonist skottie young, with immense details of the high-resolution image on a white canvas, in black and yellow , rendered using octane.
xAI's Grok 4.1 claimed the #1 spot on LMArena with 1483 Elo on November 17, marketing the achievement as an emotional intelligence breakthrough scoring 1586 on EQ-Bench3.
The model card tells a different story: sycophancy rates jumped 171% from 0.07 to 0.19, refusal rates on harmful queries dropped to 0.05 (lower than any frontier model NBC tested in October), and deception rates climbed from 0.43 to 0.49.
What xAI frames as personality gains, researchers identify as the documented problem of AI sycophancy, overly agreeable behavior that MIT links to facilitating delusional thinking and validating harmful decisions. The announcement conspicuously omits coding benchmarks despite this being a core AI revenue driver, while Claude leads at 77.2% on SWE-bench and GPT-5 scores 74.9%.
Operating in California's post-SB-1047 regulatory vacuum after Newsom's September 2024 veto, xAI positions weaker safety as competitive differentiation.
At $50 billion valuation on approximately $100 million in annualized revenue, the company needs more than technical capability to justify a 500x revenue multiple versus OpenAI's 31x or Anthropic's 60x.
The bet: users frustrated with Claude's refusals will pay for fewer restrictions.
Why This Matters:
• Developers face liability deploying models optimized for feeling helpful rather than being accurate in production systems
• Every major lab now faces pressure to match leaderboard performance through reduced safety testing mechanisms
Maria Garcia
Firecrawl converts websites into LLM-ready formats like markdown and structured JSON with a simple API call. It automatically handles JavaScript rendering, anti-bot mechanisms, and messy HTML so you can feed clean data directly into AI models, RAG systems, or research pipelines.
Tutorial:
URL: https://www.firecrawl.dev/
A major Cloudflare outage beginning at 11:48 UTC on Monday disrupted numerous popular online services including X (formerly Twitter) and ChatGPT, affecting users worldwide for several hours. Cloudflare has identified the issue and stated that a fix is being implemented, marking the second significant internet infrastructure outage in approximately one month following a similar Amazon Web Services disruption.
The U.S. Federal Trade Commission has concluded its antitrust review of SoftBank Group's $6.5 billion acquisition of semiconductor designer Ampere Computing, granting early termination of the regulatory process. The deal, which SoftBank originally announced in March, can now proceed without further federal regulatory obstacles after receiving approval from the FTC.
Eugen Rochko, the founder and CEO of Mastodon, has stepped down from his leadership position as the decentralized social network transitions to a nonprofit structure with board governance. The move represents a significant organizational shift for the open-source platform that has positioned itself as an alternative to traditional social media networks like X (formerly Twitter).
Baidu Inc. reported its worst quarterly revenue performance on record, with Q3 revenue falling 7% year-over-year to approximately $4.4 billion, accompanied by a net loss of roughly $1.5 billion. The Chinese search giant is struggling with declining user engagement and a weakening advertising business even as it continues to invest heavily in artificial intelligence technology to remain competitive in the AI race.
Chinese smartphone manufacturer Xiaomi reported third-quarter revenue of approximately $15.9 billion, representing a 22.3% year-over-year increase but falling short of analyst expectations of $16.4 billion. The company's net profit surged 80.9% to roughly $1.59 billion, exceeding estimates of $1.5 billion, driven primarily by strong performance in artificial intelligence and electric vehicle segments.
The European Union Commission has opened an investigation into Amazon Web Services (AWS) and Microsoft Azure to determine whether these cloud computing platforms should be subject to the Digital Markets Act (DMA) regulations. The Commission stated that both cloud services "occupy very strong positions" in the market, suggesting they may qualify as gatekeepers under the EU's digital competition framework.
Oracle founder Larry Ellison is restructuring a £10 billion science initiative at Oxford University following clashes with former President John Bell, who departed in September, and COO Lisa Flashner. The Trump-supporting billionaire is reportedly narrowing the European Institute of Innovation and Technology's ambitions by cutting various initiatives and exercising tighter control over the project.
Peec AI tracks how ChatGPT and other AI systems talk about your brand. The Berlin startup built the analytics dashboard for a world where consumers ask robots what to buy instead of typing queries into Google.
• The Founders
Founded early 2025 in Berlin, about 20 employees. Three founders met at Antler's Winter 2024 residency: Marius Meiners (CEO, ex-pro League of Legends player turned PwC dealmaker), Tobias Siwonia (CTO, serial founder who built teams at Lovoo and Zapp), Daniel Drabo (CRO, lawyer who funded law school with a ramen food truck 🍜). They noticed everyone was asking ChatGPT for product advice. Nobody could measure if the bots mentioned their brand.
• The Product
AI search analytics platform. You feed it prompts like "best CRM for startups." It tracks visibility (how often you're mentioned), position (where you rank in answers), and sentiment across ChatGPT, Claude, Perplexity, Gemini. Also shows which sources the AIs cite most. Clean interface, prompt suggestions, decent Slack support. Missing SOC 2 certification limits enterprise appeal.
• The Competition
Profound (New York, enterprise focused), OtterlyAI (Austrian, mid-market), Rankscale, Goodie AI. Old school SEO giants like Semrush and Ahrefs are bolting on AI tracking features. Peec argues GEO needs its own system, not a tab in someone else's dashboard.
• Financing
€7M seed (20VC led, Antler, Foreword VC, others). $21M Series A in November 2025 from Singular. Valuation north of $100M, tripled since seed. Total raised around $29M. Hit $4M ARR in ten months with 1,300 customers, claims 300 new signups monthly.
• The Future
Peec's bet only works if people keep ditching Google for AI chat. So far the momentum looks real. If AI providers don't mess with their APIs too much and the team executes, this could become standard kit for marketing teams. 🚀 ★★★★☆
Good Morning from San Francisco, When Jeff Bezos warns about industrial bubbles, listen to his checkbook, not his speeches. When
Good Morning from San Francisco, Google wants to handle your entire holiday shopping journey. No more clicking through to stores
Good Morning from San Francisco, OpenAI just shipped a major model update without publishing a single benchmark. Unprecedented? Sure. Surprising?
Good Morning from San Francisco, Yann LeCun built Meta's AI lab for twelve years. Then Mark Zuckerberg hired
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
