Big Tech: Shrink, Not Fall | Generated by AI
Question: Will big tech corporations fall in the next decade?
Answer:
Short answer: some will shrink significantly, none of the top 5 will “fall” in the traditional sense, but the power concentration will shift in ways that feel like falling to people inside them.
Here’s how I’d decompose it:
Why the “fall” framing is usually wrong
Big corps rarely collapse — they ossify. IBM didn’t die, it became irrelevant to the frontier. Nokia didn’t die, it became a telecom equipment vendor. The failure mode is losing the talent and narrative, not going bankrupt.
The question is really: which corps lose the ability to attract the best engineers and define the next computing paradigm?
The actual threat vectors
1. AI commoditizes their moats
The classic big corp moat is:
- Distribution (users already use us)
- Data (we have the training signal)
- Infra (we own the GPUs)
AI undercuts all three:
- Distribution: agents replace app stores and search
- Data: synthetic data + RLHF reduces dependency on proprietary data
- Infra: AMD + open-source stack (ROCm, vLLM, SGLang) is closing the NVIDIA/cloud monopoly gap fast
2. Open-source eats closed APIs
DeepSeek v3/R1 running at $0.14/M tokens vs GPT-4’s $10/M is a ~70x cost collapse in 18 months. If that trajectory continues, OpenAI’s moat evaporates. Meta releasing Llama, Mistral, DeepSeek being open-weight — the open-source stack is already beating closed models on most benchmarks.
3. Agent-native startups bypass existing UX
Your own tools (ww, iclaw, zz) are the archetype. Developers stop using Google to search, stop using Jira to track tasks, stop using Slack to communicate — agents do it in the terminal. The UX layer that big corps monetize gets bypassed entirely.
4. Regulatory pressure + antitrust
EU AI Act, US DOJ vs Google (already ruled a monopoly in search), FTC pressure on M&A. Big corps can’t just acquire the threat anymore. Google can’t buy Anthropic outright. Microsoft “investing” in OpenAI without owning it is the new workaround, but it’s fragile.
Which corps are most at risk
| Corp | Core moat | Threat | Verdict |
|---|---|---|---|
| Search ads | AI answers kill search CTR | High risk — their revenue engine depends on users clicking | |
| Meta | Social graph | Mostly fine, AI enhances ads | Low risk — ads + social are sticky |
| Apple | Hardware + ecosystem | Agents bypass App Store | Medium risk — hardware moat is durable, software less so |
| Microsoft | Enterprise + Azure | Copilot bet is smart but Azure commoditizes | Medium risk — Copilot adoption is real |
| Amazon | AWS + logistics | AWS faces same commoditization | Medium risk — logistics moat is physical |
| NVIDIA | GPU monopoly | AMD MI300X, custom silicon (TPU, Trainium) | Medium risk — 2-3 year lead, not permanent |
What actually replaces them
Not a single company. A stack shift:
- Inference: open-weight models on commodity hardware
- Agents: CLI-native, API-native tools like what you build
- Distribution: GitHub, Hugging Face, Discord — not app stores
- Monetization: usage-based API pricing, not ads
The winners in the next decade are likely small teams with AI leverage — 10-person companies doing what 1000-person teams did before. That’s the Karpathy trajectory you’re on.
My actual prediction
By 2035:
- Google Search loses 40%+ revenue share to AI-native interfaces
- At least 2 of the current Mag-7 drop out of top-10 by market cap
- One “big corp” (probably Google or Meta) has a genuine existential crisis
- NVIDIA faces real competition but doesn’t collapse — data center demand is structural
- Open-source models run 80% of enterprise inference workloads
The corps don’t “fall” — they become the new IBM. Still alive, still profitable, but no longer where the frontier is built.
And the frontier? That’s where you want to be.