Remember when everyone counted Google out of the AI race?
In early 2023, Google's narrative was in the gutter.
OpenAI had ChatGPT. Microsoft had the partnership. Google was “asleep at the wheel” and about to lose search to AI.
Fast forward to today. Gemini has 650M+ users. Google’s stock is at all-time highs. They caught up… because they had the raw ingredients all along: data, distribution, compute.
I’m starting to wonder if the same dynamic is playing out with Meta right now.
The market hates Meta’s AI story.
Stock dropped 12% after the last earnings call. Wall Street sees billions in capex with “nothing to sell.” Llama 4 got a lukewarm reception. The bear case writes itself.
But here’s what I’ve been noodling on:
- Meta has 3.4B users across its apps (instant distribution)
- They’re building a 5 GW data center (unprecedented compute)
- They just assembled an AI “dream team” including Scale AI’s founder as Chief AI Officer
- Llama is already the foundation for most open-source AI development
Sound familiar? Data + distribution + compute. The same ingredients Google had.
So I framed my hypothesis: Can Meta realistically close the frontier AI gap within 12 months? And if so, what would that mean for the stock?
I’m going to show you how I used ChatGPT Deep Research to build a full competitive intelligence dossier on Meta vs frontier AI labs.
Here’s the plan:
- Frame the core research question and sub-hypotheses
- Run a deep research prompt to build a 50+ page dossier
- Extract the key findings that actually matter for the investment thesis
The Prompt
We're using a 2 part prompt to craft this deep research: (1) A prompt to craft the prompt (2) Running the Deep Research prompt
I find ChatGPT Deep Research best for this type of full length dossiers.
Prompt 1:
You are an expert prompt engineer specializing in deep research prompts for investors. I need your assistance crafting a deep research prompt. For background information, I am an equities investor looking into Meta and determining the research question "Can Meta catch up for AI? Can Meta eventually compete with a frontier AI stack or business model?"
I have these questions below that I want to investigate but I am sure you have research questions that are relevant that I did not include - I am open to including your research questions as well.
1. Lay out the main "bearish" concerns the market has for Meta AI for me to understand the situation.
2. What are the AI assets Meta has? Data? Distribution? Reach? Louisana data campus?
3. What is meta's AI angle that differentiates vs the other big model players like OpenAI and Google? This can include monetization mechanism, model building philosophy, etc.
4. Research into academic papers that Meta has published .... what are they doing in the published paper research side? Have they made breakthroughs?
5. Research into the key leadership and key AI research roster working for Meta
6. Timeline for Meta's AI releases this year
Before you begin please ask me any clarifying questions you may have.
Prompt 2: dave_wang_meta_deep_research_prompt.pdf
The Result
Full output here: Link
Our AI deep research produced a 50+ page dossier covering Meta's assets, constraints, competitive positioning, timeline, and scenario analysis.
Here are the findings that actually matter:
The Capability Gap is Real... But Narrowing
- Llama 4 still trails GPT-4/5 on key benchmarks like MMLU and complex reasoning
- But Meta's MoE architecture reportedly runs at 1/20th the cost of GPT-4 for similar tasks
- Their Perception Encoder achieved state-of-the-art on zero-shot image/video classification... beating both open AND proprietary models
- Spring 2026 target for "Avocado" (their closed frontier model) to reach GPT-5 parity
The Raw Ingredients Are There
The Bear Case Has Merit
- No direct AI revenue yet (unlike OpenAI's subscriptions or Google Cloud)
- $48B in property & equipment spend in first 9 months of 2025 (216% YoY)
- Stock dropped 12% after signaling even higher spend without clear monetization plan
- Open-source strategy might commoditize the very thing they're trying to build
The Bull Case is Underappreciated
- Meta doesn't need to sell AI directly... they can monetize via engagement and ads
- Open-source could be the "Android play" for AI (ubiquity over margin)
- Hybrid strategy emerging: open-source for ecosystem, closed models for enterprise revenue
- Google memo few years ago literally said "we have no moat, and neither does OpenAI"... largely because of Meta's Llama releases
Key Milestones to Watch (Next 12 Months)
- Spring 2026: Avocado model launch (the make-or-break moment)
- Hyperion Phase 1 coming online
- Any enterprise deals or API revenue announcements
- Benchmark results vs GPT-5 and Gemini 3
My Take
The AI deep research surfaced a nuanced picture. Meta can plausibly close the capability gap... but execution risk is high.
Meta's constraint isn't technical capability. It's monetization clarity (But that is something Meta has solved in the past with socials biz!). They have the ingredients. The question is whether they can cook.
This brings me to a bigger point about how investing is changing.
With AI, the bottleneck is no longer gathering information. I compressed what would've been weeks of competitive analysis into a single afternoon.
The new bottleneck? Asking the right questions.
Your job as an investor is shifting from "information gatherer" to "hypothesis tester." The edge now comes from developing the gut feel to frame the right research questions... then letting AI stress test them.
I framed a hypothesis. AI did the heavy lifting. Now I have a view to refine.
I'm going even deeper on this one.
The Meta AI question is too interesting to leave at a single newsletter prompt. I'm building out a full due diligence series on whether Meta can catch up to frontier AI... and what it means for the stock.
I'll be publishing the full analysis on and . If you're not following me there yet, now's the time.
Hit reply if you have specific angles you want me to dig into.
Personal:
I'm spending the next month in Asia to meet with our investment bank and hedge fund clients!
Hit the reply button if you're around. Maybe I'll organize an event if there are enough folks around...