(Preview) Meta and Its Messaging Problem, The XBOX Reset, Q&A on Token Costs, American Soccer, Starlink in Nature

Summary

• Ben Thompson’s investment case is that Meta may have one of the clearest AI monetization paths among companies pursuing frontier models because it already has consumer reach, data, and advertising machinery. LLMs need not achieve AGI: Meta’s digital ad system serves trillions of ads a year, and if its ads succeed only 0.2% of the time, that can still make it the best ad platform on Earth. “They just need to build the engine here.”
• Meta’s CapEx and hiring are not the problem, although Llama 4 suggests execution has been uneven. Thompson says Llama 2 and Llama 3 seemed solid before something “totally went off the rails,” but AI remains existential for a purely digital company with numerous ways to earn a return.
• Thompson’s real charge against Mark Zuckerberg is a “sin of omission”: Zuckerberg has not fully embraced Facebook as “the greatest ad company of all time.” That failure produces muddled investor messaging, can create avoidable morale damage during stock drawdowns, and has contributed to a recurring “desire to be something that Facebook isn’t,” including spending in the wrong direction.
• The historical costs of that confusion include hardware, the metaverse, and Meta’s slow recognition that Facebook is entertainment rather than merely a friends-and-family network. Thompson argues hardware inherently constrains a company built to reach everyone, while Facebook’s reluctance to embrace lean-back entertainment let TikTok build itself “on Facebook, under Facebook’s noses.”
• Muse-Spark-1.1 illustrates the difference between sensible internal technology and incoherent public positioning. Meta needs coding models for its own development and is reportedly Anthropic’s biggest customer, paying it billions annually, while agentic image generation could transform ad creation. But Zuckerberg making a coding model the subject of his first tweet since 2023 was, in Thompson’s words, “asinine.”
• Meta should invest aggressively in AI without presenting itself as another OpenAI or Anthropic. Thompson’s analogy is that Zuckerberg keeps leaving an ocean Meta already dominates for a “shark-infested lake”; personal superintelligence or a personal assistant might eventually emerge from Meta’s consumer data, but the immediate, credible story is better ads, better content prediction, entertainment, and human connection. “Just talk about it. Be happy about it. Embrace it.”

Deep dive

1. Zuckerberg’s messaging problem begins with Meta’s identity

• Thompson described himself as positive on Meta’s opportunity but said he could not fully make his usual bottom-of-cycle declaration: “Everyone outside is stupid. Obviously this is a great investment.” The repeated leadership pattern finally had to be addressed.

• His diagnosis is a “sin of omission,” not primarily Zuckerberg’s spending decisions: Zuckerberg has not fully embraced what Meta already is, so he has not delivered the messaging that would give him freedom to do what has to be done. The underlying frustration is “this desire to be something that Facebook isn’t.” Thompson said that omission has also contributed to sins of commission, including spending in the wrong direction.

2. Meta does not need AGI for its AI economics to work

• Thompson said the model record is mixed: Llama 2 and Llama 3 seemed solid, then “something totally went off the rails with Llama 4.” He believes many Mistral employees were former Llama people and called Yann LeCun a legend but “totally the wrong person for Meta” in the LLM area.

• LeCun’s skepticism about LLMs reaching AGI is beside the point for Meta. Advertising already operates probabilistically: baseball greatness means succeeding one-third of the time, while an ad platform whose ads succeed around 0.2% of the time can be the world’s best.

• That is why Thompson has no objection to the CapEx or hiring. Meta is a resolutely consumer company and should not become a B2B company; AI is existential for a purely digital company, and Meta has many ways to apply it. Unlike frontier labs whose returns he said remain unclear, Meta already has “all the mechanisms to earn a return in place.”

3. Fighting the ad business created Meta’s strategic detours

• Thompson’s preferred foundation is categorical: “Facebook is the greatest ad company of all time,” and its ads can be a societal good. Lean-back feeds are unusually fertile because users are not doing serious work: an ad is simply more content interrupting another Instagram Reel.

• Platforms feature third parties; advertising features advertisers. Thompson sees that conflict behind Meta’s hardware ambitions: a service intended for everyone should not restrict itself to devices it sells, yet Facebook has pursued hardware since 2012 and, in his telling, spent hundreds of billions of dollars building a hardware business. He contrasted this with Apple delivering the smartphone world to Facebook “on a platter,” then argued that Apple got away with its ATT policy because Facebook did not effectively speak up; he called ATT one of the most egregious antitrust actions a company has taken.

• The same identity problem delayed Meta’s entertainment pivot. Thompson says he argued in 2015 or 2016 that Facebook was more than friends and family; had it accepted that framing then, it would have recognized TikTok earlier instead of letting TikTok build itself “on Facebook, under Facebook’s noses.”

4. Muse-Spark is useful technology wrapped in the wrong story

• Andrew Sharp pointed to Zuckerberg’s first tweet since 2023, announcing Muse-Spark-1.1, described as a strong agentic encoding model offered through the new Meta Model API and Meta AI. Thompson dismissed making it the public focus—“Who cares?”—while saying building the model itself makes sense.

• Meta is reportedly Anthropic’s biggest customer and pays it billions of dollars annually, so internal coding capability is strategic. Agentic image generation also maps directly to ads: a prompt could search the web for reference images, reason, use tools, and iteratively create Andrew with Trae Young’s hair. “Framing yourself as a coding company is asinine.”

5. Meta should defend its ocean instead of chasing frontier labs

• Andrew pushed back that Zuckerberg’s persistence is not always redeeming: Reality Labs went in the wrong direction, Meta literally renamed the company after the metaverse, and, in Andrew’s words, that is now a “giant flop” hanging over everything.

• Thompson said many investors would prefer harvesting cash and returning it through buybacks, but compared Meta’s frontier-model ambition to aspiring newsletter writers competing directly with Stratechery. A better strategy is: “Find your own pond.”

• Meta already has the largest lake—or an ocean of its own—in consumer entertainment, connection, data, and advertising, whose value is being enhanced by AI, yet keeps trying to “flop onto land” toward OpenAI and Anthropic. Meta’s consumer data could eventually support a personal assistant or even personal superintelligence, but Thompson says that is not the vision to lead with: most people want to sit back and be entertained, and the immediate line from AI to substantially better ads is already clear, including predicting the next ad rather than merely matching one and eventually predicting content too.