Meta launches Muse Spark 1.1, plans own AI chip production

Meta Platforms announced two significant AI-related moves on July 9, 2026. First, the company launched Muse Spark 1.1, an upgraded version of its AI model that improves on version 1.0 in two key areas. The most notable upgrade is the context window, which expands to 1 million tokens—roughly four times larger than the prior version. A larger context window allows an AI agent to work on a task for longer without losing track of what it is doing, addressing a real problem where AI agents lose context mid-task and continue consuming resources. Equally significant is pricing: Muse Spark 1.1 costs more than 50% less than competitive offerings from Anthropic and OpenAI for both input and output, a decisive advantage in an AI market dominated by pricing competition. Meta can afford this aggressive pricing because its core advertising business—which generates far higher revenue per user than competitors like Pinterest or Snap—funds AI development as a secondary effort.

Second, Meta announced plans to manufacture its own AI chips, with production starting in September. Broadcom will help design the chips and Taiwan Semiconductor Manufacturing Company will manufacture them—the same model Apple uses for iPhone chips. The goal is to reduce Meta's dependence on expensive Nvidia and AMD processors as the company builds out its compute infrastructure. Meta plans to double its compute capacity next year, making custom chips an attractive cost-reduction lever. However, Meta will continue to rely on Nvidia and AMD for the most demanding AI tasks and has signed what is described as the biggest AI deal in history to purchase AMD chips. Matt Frankel noted that this approach mirrors what Google and Amazon have done for years: custom chips handle easier workloads while premium off-the-shelf processors handle the hardest problems.

Contributors debated whether these moves represent disciplined strategy or unfocused sprawl. Tyler Crowe expressed skepticism, framing Meta as a company making many bets—creative financing for data centers, smart glasses, prediction markets, and AI models—without a clear direction, ultimately relying on its core advertising business as a fallback. Jon Quast countered that Meta's financial performance justifies the approach: operating income grew 20% in 2025 and accelerated to 30% in 2026, suggesting gains come not only from advertising but from overlapping investments in AI. He noted the stock trades at 22 times earnings while growing at this pace, making it cheaper than the overall market. The podcast was recorded on July 9, 2026.

Meta's announcements reflect a company pursuing multiple AI infrastructure strategies simultaneously. The Muse Spark 1.1 pricing undercuts established competitors because Meta can afford aggressive pricing—unlike Anthropic and OpenAI, whose business models depend entirely on selling AI services. Meta's advertising operation generates the cash flow that allows it to invest heavily in AI as a secondary initiative. The custom chip production, meanwhile, follows the template already proven by Google and Amazon: design chips to handle routine workloads (with partners like Broadcom and Taiwan Semiconductor) while continuing to buy premium processors from Nvidia and AMD for harder tasks. This dual approach aims to stretch compute capacity further as Meta plans to double its capacity next year. However, contributors on the podcast disagreed on whether these moves signal strategic discipline or unfocused sprawl. Tyler Crowe criticized Meta for "throwing spaghetti at the wall," noting past failed attempts at secondary revenue streams like the metaverse. Jon Quast countered that Meta's operational income growth—20% in 2025 and 30% in 2026—demonstrates real returns from AI investment, and at 22 times earnings, the stock trades cheaper than the broader market despite rapid growth.