🧠 Google just fused Gemini into silicon

Plus 36 non-AI stocks Goldman likes, AMD's Nvidia rival ships, and the US AI safety chief is already gone.

Author

Anthony Chebat
July 20, 2026

Sponsored by

My fellow AI explorers

Markets did something weird to me this week: I found myself more interested in everything happening with a chip that doesn't exist yet than in the models that do.

This edition's a market mover.

In today’s edition:

The best voice models, now across all channels

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

AI Chips

🧠 Google Isn't Just Building a Better Model. It's Building the Model Into the Chip.

Alphabet shares jumped as much as 3% on Monday after Google reportedly started developing a new server chip, internally called Frozen v2, that could ship as soon as 2028. The headline is efficiency: the chip is projected to run six to ten times more efficiently than Google's current TPUs.

Here's what stood out:

  • It's not a TPU. Frozen v2 sits alongside, not inside, Google's existing tensor processing unit lineup, meaning Google is now running parallel hardware bets rather than one roadmap.

  • Gemini lives in the silicon. Rather than general-purpose hardware running Gemini as software, the chip's architecture reportedly embeds pieces of Gemini's model design directly into it, cutting the data movement that slows everything down.

  • The timing isn't subtle. This drops two days before Alphabet reports Q2 earnings on July 22, with investors already bracing to see whether $180 to $190 billion in 2026 AI capex is paying off.

Here's the asterisk: for years, the AI hardware story has been "who has the most chips." Google's move suggests the next fight is "whose chip only runs your model well." If Frozen v2 is genuinely model-specific silicon, Google is making a bet that vertical integration, not raw compute, wins the next round. That's a very different game than the one Nvidia has been winning.

🔮 Prediction: Expect Amazon and Microsoft to accelerate their own custom silicon roadmaps in response. The "everyone buys Nvidia" era isn't over, but the "everyone builds their own chip too" era just got a serious green light from the most cautious of the hyperscalers.

Reply and tell me: Does model-specific silicon make you more or less bullish on Nvidia's moat?

Chinese AI

📉 Moonshot's Kimi K3 Just Wiped Out $314 Billion, and It Isn't Even Public Yet

China's Moonshot AI released Kimi K3 last Thursday, a 2.8 trillion-parameter model the company says rivals top-tier offerings from OpenAI and Anthropic at a fraction of the cost. Markets reacted like it was DeepSeek all over again.

Here's what stood out:

  • The number that matters: IG analyst Tony Sycamore estimates $314 billion has been wiped from private valuation estimates for OpenAI and Anthropic alone, despite neither company being publicly traded.

  • It beat models that don't exist for the public yet either. Per Axios, Kimi reportedly beat Anthropic's Fable 5 and OpenAI's GPT-5.6 in front-end coding tests, and outranked Anthropic's prior flagship Opus 4.8, all at roughly 40% lower cost.

  • The weights aren't even out. Moonshot won't release Kimi K3 as open-weight until July 27. Every headline this week is reacting to benchmark claims and demos, rather than to code anyone outside Moonshot can independently verify.

Here's the asterisk: the market didn't wait for confirmation, and neither did Washington. Axios also reports the Trump administration is weighing new restrictions on Chinese open-weight models, even as US companies increasingly rely on exactly those models because they're cheap and good.

You can't credibly fear China catching up while also blocking American developers from using the thing that just caught up.

🔮 Prediction: If the July 27 weight release backs up the benchmark claims, expect a second, sharper leg down in AI infrastructure valuations, and a much louder push in DC for open-source restrictions that most technologists will call self-defeating.

Reply and tell me: Should the US restrict access to Chinese open-weight models, or does that just slow America down too?

30-Second AI Play

💰 Goldman's Non-AI Stock Screen, In One Place

Tired of every "smart money" list being another chip stock? Goldman Sachs strategist Ben Snider just published a screen of 36 stocks the bank believes are insulated from AI disruption because they're built around physical, in-person experiences an AI agent literally cannot attend for you. The group has returned 17% year to date, versus 11% for the equal-weighted S&P 500.

  1. Movies and entertainment: Walt Disney, Live Nation, Liberty Media (Formula One), TKO Group, Madison Square Garden Sports, Sphere Entertainment, Atlanta Braves Holdings, Madison Square Garden Entertainment.

  2. Hotels, resorts, and cruise lines: Marriott, Royal Caribbean, Hilton Worldwide, Viking, Carnival, Hyatt, Norwegian Cruise Line, Wyndham, Choice Hotels, Travel + Leisure, Hilton Grand Vacations, Marriott Vacations Worldwide.

  3. Casinos and gaming: Las Vegas Sands, MGM Resorts, Wynn Resorts, Boyd Gaming, Caesars, Churchill Downs, Red Rock Resorts, PENN Entertainment, Monarch Casino & Resort.

  4. Consumer services and leisure: Liberty Live, Life Time Group, Vail Resorts, Planet Fitness, OneSpaWorld, United Parks & Resorts.

  5. Do the homework now. Snider's note flags Disney, Royal Caribbean, Carnival, Las Vegas Sands, Norwegian Cruise Line, Boyd Gaming, Churchill Downs, Wyndham, Choice Hotels, Planet Fitness, Marriott Vacations Worldwide, United Parks & Resorts, and Travel + Leisure as trading below the S&P 500's price-to-earnings multiple, meaning there's still value before the "non-AI trade" becomes its own crowded bet.

💡 Pro tip: The median stock on this list trades at 12 times forward EBITDA, which Goldman says sits in a historically reasonable range. This is a homework list, not a buy list. Do your own diligence before touching any of it, and remember Claude isn't a financial advisor.

Advertise to 180k engineers and CTOs choosing what tools their companies build with.

Other Relevant AI News!

🖥 AMD finally ships its answer to Nvidia. The company's Helios rack-scale AI system is landing at Microsoft, joining Meta, OpenAI, and Oracle as customers, and CNBC got an exclusive first look inside the Texas lab where it's built. Nvidia still controls over 95% of the data center GPU market, but this is the most credible rival system in years.

💸 A $15 million bet on cracking Nvidia's software moat. Inference infrastructure startup Infinity, founded by former Google Brain researcher Jeremy Nixon, raised $15 million from Touring Capital, Principal VC, and researchers from OpenAI and Anthropic, aiming to make it easier for non-Nvidia chips to actually run AI models.

🌐 Washington's China AI dilemma just got awkward. Axios reports the Trump administration is quietly weighing curbs on foreign open-weight models, even as US firms lean harder on cheap Chinese models like Kimi. Pro-competition voices inside the industry warn a ban would hand China the open-source lane outright.

🏛 The federal AI safety chief is already gone. Chris Fall has resigned as director of the Commerce Department's Center for AI Standards and Innovation, just three months into the job, with NIST's Arvind Raman stepping in temporarily. No reason was given for the departure.

Golden Nuggets

  • 🧠 Google is betting the next chip war is won by hardware built for one model, not every model.

  • 📉 A model nobody can verify yet already erased $314 billion in paper value, and Washington is scrambling to respond.

  • 💰 Goldman's "physical experience" stock screen is quietly beating the AI trade this year.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI