Overview

In this final installment of a 15-episode deep dive into Meta (formerly Facebook), hosts Kazuhiro Obara (an IT critic and former McKinsey/Google executive) and Kensuu (serial entrepreneur and CEO of Aru Inc.) dissect Meta's AI strategy with striking clarity. The central thesis is that Meta is deliberately pursuing an open-source AI model (Llama) not out of altruism, but as a calculated competitive move to prevent any single rival—particularly OpenAI—from becoming a monopolistic "one strong player" in AI. By making its AI freely available and encouraging developers to build on it, Meta shifts the battlefield from pure model performance to data ownership, where Meta's unparalleled trove of social graph data (10x Google's search data volume) gives it an insurmountable advantage. The conversation feels like a masterclass in strategic thinking, blending historical analogies (Android vs. Apple, Windows vs. Mac) with forward-looking analysis of how Meta plans to embed AI into everyday life through WhatsApp, Instagram, and Facebook.

---

0:05Meta's Late but Deliberate Entry into AI

Meta's serious pivot to AI began only in February 2023—remarkably recent compared to its VR investments, which started in 2014. Obara notes that when Facebook laid out its "10-year plan" back in 2016, it identified three pillars for the post-2016 era: connectivity (bringing the last 3–3.5 billion unconnected people online), VR/metaverse, and AI. But AI remained an abstract concept until the release of Llama in February 2023. The hosts emphasize that this timing was not accidental—it was a strategic response to the landscape at that moment.

The core logic, as Obara explains, is brutally simple: "If you can create a structure where your rivals cannot win, then Meta—which has the most data—will win." This is the essence of what he calls a "judo strategy": using an opponent's strength against them. OpenAI had become the dominant force in AI by early 2023, with Google in "code red" mode and even co-founder Sergey Brin returning to active involvement. But OpenAI, despite its name, had become increasingly closed—stopping publication of technical papers after GPT-3.5, restricting API access, and becoming what critics called "ClosedAI." This created an opening for Meta to position itself as the open alternative.

---

2:28Open vs. Closed: The Historical Framework

Obara introduces a powerful historical framework to explain Meta's thinking: the battle between open and closed platforms. He points to smartphones as the canonical example. Android (open) won in terms of global market share—roughly 70% of devices run Android versus 30% for iOS. Yet Apple (closed) commands a higher market capitalization despite selling fewer than half the units. The lesson: closed models can capture more value per user, while open models win on scale and ecosystem breadth.

The hosts trace this pattern further back. In personal computers, Microsoft's Windows (an open model where any manufacturer could build compatible hardware) defeated Apple's closed Mac ecosystem. In each case, the open model won the volume game, while the closed model captured premium profits. The question for AI becomes: which path will dominate?

Obara argues that Meta is betting on open winning in AI, just as Android won in smartphones. By releasing Llama as an open model, Meta aims to make its technology the default foundation for the entire AI ecosystem. The strategy is to commoditize the AI model layer—making high-quality AI freely available to everyone—while Meta itself profits from the data and distribution advantages that no competitor can replicate.

---

4:52The Llama Strategy: From Research Tool to Industry Standard

The timeline of Llama's evolution reveals a carefully orchestrated rollout. When Llama 1 was released in February 2023, it was initially a research-only model—available for academics and researchers to experiment with, but not for commercial use. This was a clever entry point: at a time when OpenAI was closing off its models, Meta offered researchers a lightweight model that could run on a single GPU (roughly 6.7 billion parameters), meaning it could operate on a standard consumer PC.

Just five months later, in July 2023, Meta released Llama 2 with a critical change: it permitted commercial use. Even more importantly, Meta allowed developers to use Llama to train other AI models—essentially giving permission for the entire AI community to build on top of Meta's foundation. The result has been staggering. On Hugging Face (the GitHub of AI, where researchers share models), over 500,000 AI models have been posted, and an estimated 80% of independent AI models are now trained on a Llama base.

Obara emphasizes that this is the "real joy of the open model strategy": winning over developers. By making Llama the default starting point for AI development, Meta ensures that its architecture becomes the de facto standard. The upcoming Llama 4, while not yet released, is reportedly aiming to surpass ChatGPT in scale while remaining freely available—potentially enabling applications like climate change simulation and drug discovery that require massive computational resources.

---

10:07Meta AI: 1 Billion Users and the Personal AI Vision

While Meta's open-source strategy targets developers, the company has simultaneously built a consumer-facing product called "Meta AI" that has already reached 1 billion monthly active users—roughly the same milestone that ChatGPT recently announced. The hosts note that Meta AI is not yet available in Japan, but its global reach is enormous, particularly in emerging markets.

The key insight is that Meta is not trying to build the most powerful AI in the world. Instead, it is building what Obara calls a "personal AI"—an assistant deeply integrated into users' daily social lives. Kensuu points out that Meta's unique advantage is its access to the social graph: Facebook, Messenger, WhatsApp, and Instagram data. This allows Meta AI to answer questions like "What was my friend obsessed with last time we met?" or "Which brand does my friend like?"—queries that require understanding personal relationships and social context.

Obara illustrates this with a concrete example: when planning a trip, younger users increasingly turn to Instagram rather than Google for recommendations. They want to know what influencers they follow are doing, what events their favorite celebrities are attending, and what brands their friends are liking. Meta AI can provide exactly this kind of personalized, socially-aware guidance because it has access to the underlying social data. This is fundamentally different from ChatGPT, which is a general-purpose knowledge engine but lacks personal context.

---

15:50Data as the Ultimate Moat: Why Meta's 10x Advantage Matters

The hosts pivot to what they consider the most critical competitive factor in AI: proprietary data. Obara cites research from Scott Galloway at NYU that quantifies the data advantage of major tech companies. Using "tokens" (the basic units AI models process) as a metric, the analysis shows that Google's search data is substantial, but X (formerly Twitter) actually generates 1.5x more data than Google Search because of its real-time, high-volume posting.

But Meta's data dwarfs both. The combined messages and posts across Facebook, Instagram, and WhatsApp amount to roughly 10x the data volume of Google Search. This is not public data that any AI can crawl—it's private, closed data that only Meta can access. Kensuu makes the point explicit: "Google can search Meta's data? No. X's AI can search Meta's data? No. So the closed data that only you have—that's what creates competitive differentiation in AI."

This data advantage becomes decisive as AI model performance converges. In 2023, OpenAI was the clear leader. But by 2024–2025, models like Claude, Gemini, and Grok have all become competitive. When model quality is roughly equal, the winner is determined by who has the best data. Meta's 10x data advantage means it can train more personalized, more contextually aware AI than any competitor. Conversely, OpenAI has almost no proprietary user data—which is why Sam Altman is desperately trying to build products where users will "deposit their life data," including the reported ~$1 trillion acquisition of Jony Ive's design firm to create a new hardware product.

---

20:01Live Data vs. Stock Data: The Next Frontier

Obara introduces a crucial distinction between types of data. "Stock data" is historical data that already exists—past posts, old messages, archived content. "Flow data" is real-time activity. But the most valuable category, he argues, is "live data"—data about what a person is about to do next, their intentions and plans.

Google has some live data through Google Calendar and Google Maps (where you're going, what you're planning). But Meta has something even more intimate: data about who you're meeting, what you're thinking about, and your social intentions. This is the data that comes from messaging apps and social platforms—the "who are you meeting tomorrow?" and "what are you excited about?" type of information.

This is why Meta's strategy of keeping AI open while keeping data closed is so powerful. By ensuring that no single AI model becomes overwhelmingly superior (through open-sourcing Llama), Meta prevents a scenario where a competitor's superior AI could overcome Meta's data advantage. If all models are roughly equal, the one with the best data wins—and that's Meta. Kensuu summarizes it elegantly: "If OpenAI alone had overwhelmingly better performance, no matter how much data you have, you'd lose on performance. But if everyone's performance is about the same, it becomes a data battle—and Meta is overwhelmingly advantaged."

---

23:27Rule-Making as the Ultimate Strategy

The episode concludes with a broader reflection on what makes Meta's approach so instructive for entrepreneurs. Obara argues that the most powerful strategic move is not to compete within existing rules, but to change the rules of the game itself. Meta's AI strategy is a textbook example: instead of trying to beat OpenAI at building the best closed model, Meta changed the game by making AI open, thereby shifting the competitive axis from model quality to data ownership.

This "rule-making" mindset, Obara suggests, is what separates great companies from good ones. It requires seeing beyond the current competitive landscape to imagine what the next battlefield will look like—and then designing the terrain to your advantage. For entrepreneurs, this means asking not "how do I win in this market?" but "how do I change the market so that I naturally win?"

The hosts tie this back to earlier episodes in the Meta series, noting that the same strategic thinking applied to Meta's VR/metaverse investments and its Libra cryptocurrency project. In each case, Meta was trying to define the next platform rather than just compete on the current one. Kensuu reflects that this perspective helps entrepreneurs see "where the next growing market will be" and "what holes exist in the current market"—allowing them to multiply their efforts by riding market tailwinds rather than fighting headwinds.

---

Conclusion

What stays with the listener is the elegant simplicity of Meta's AI strategy: open the model, own the data, and embed AI into the social fabric of daily life. The episode succeeds because it doesn't just describe what Meta is doing—it explains *why* it makes strategic sense, using historical analogies and data-driven analysis. The hosts' enthusiasm is infectious, particularly when they connect Meta's moves to broader lessons about entrepreneurship and competitive strategy. This episode matters because it offers a framework for thinking about platform competition in the AI era that applies far beyond Meta itself. Whether you're an entrepreneur, investor, or just a curious observer, understanding why Meta chose openness over closedness—and how data becomes the ultimate moat—provides a lens for evaluating every major tech company's AI strategy going forward.

---

Key takeaways

• Meta's AI strategy is a "judo move": by open-sourcing Llama, it prevents any single competitor (especially OpenAI) from becoming dominant, shifting competition to data ownership where Meta has a 10x advantage over Google.
• The open vs. closed platform dynamic has historical precedent: Android (open) won market share, Apple (closed) won profits—Meta is betting open wins in AI just as it did in smartphones.
• Llama has become the de facto standard for independent AI development, with ~80% of non-major AI models built on a Llama base, hosted on Hugging Face.
• Meta AI has already reached 1 billion monthly active users, matching ChatGPT's milestone, by focusing on "personal AI" integrated into WhatsApp, Instagram, and Facebook rather than general-purpose knowledge.
• Meta's proprietary data advantage is staggering: its combined social and messaging data is 10x the volume of Google Search data, and this data is private and inaccessible to competitors.
• The most valuable data is "live data"—what people are about to do—and Meta's social graph gives it unique access to users' intentions and social plans.
• As AI model performance converges across competitors, the decisive factor becomes who has the best proprietary data, making Meta's position nearly unassailable.
• The ultimate strategic lesson is "rule-making": instead of competing within existing rules, change the game itself to favor your strengths—Meta did this by making AI open when everyone else was closing it.