Agentic AI is framed as a huge infrastructure tailwind while threatening ad layers, especially AMZN’s, with the offset that compute demand could more than replace the lost ad dollars over time. The key investor problem is where value lands in agentic shopping: AMZN, SHOP, AAPL, and upstream power names could benefit, but each layer can also be disintermediated by agents. AMZN’s internal chips, shopper data, and Moonraker/Alexa work are presented as its defenses, while AAPL’s on-device assistant model is preferred for privacy and control. BLOOM is cited as a clear beneficiary because compute still needs electricity regardless of which agent stack wins.
Genetic AI is expected to be the biggest positive ever for infrastructure companies because agents will burn massive compute through simultaneous token runs.
Genetic AI could crush the advertising layer as it exists, and that layer is described as a big part of AMZN’s earnings.
Any ad loss is expected to be offset over time by 2x-3x more infrastructure spend.
AMZN has a shot at building an agent-to-agent advertising model for transactions and purchase decisions, including deciding whether paid products get gating or preferential treatment.
AMZN is building its own agentic broad-knowledge bot to improve Alexa, with project Moonraker described as an internal code name for a full agentic Alexa that can design and execute complex multi-step tasks from one request.
AMZN is not dependent on NVDA or other outside chip suppliers because it is building internal chips.
AMZN’s shopper data is described as the most extensive in the world.
Agents are expected to take over more purchase decisions and transactions, but not quickly.
Products like Moonraker are expected to proliferate into roughly 10 similar agent products across tech companies.
Multiple agentic personal assistants reading email at once create confusion because users cannot tell why each one flags different things.
A connector that syncs assistants’ knowledge is proposed so one assistant can continue context across other assistants.
Giving an agent email access is described as risky; one example given is Instagram shutting down access after Muse got into Instagram DMs.
Having multiple agentic assistants reading and acting on email right now is described as too much.
AAPL is preferred to win because assistant functions could stay on-device instead of in cloud or private cloud compute; if Siri can read Apple Mail and iMessage, those are the main communication channels, while Gmail is just an address given to companies.
Current Siri is described as doing summarization, while a new Siri update to iOS 27 is claimed to add actions like putting calendar items onto the screen.
Agentic shopping is described as certainly coming and 'going to be huge,' making it the main investor issue.
SHOP is framed as an obvious beneficiary of agentic shopping, but the question is whether SHOP is still needed in an AGI/ASI world.
Each agent could clone its own Shopify, making it unnecessary for agents to share value with SHOP.
It is hard to tell where value lands in an age of agentic shopping, which is why stock prices are described as moving erratically day to day.
BLOOM is called a likely winner because it sits so far upstream that it should benefit no matter who wins the AI stack, since electricity is still required to power compute.