🤖 A CUDA transpiler breaks Nvidia's software lock-in
Chamath argues the real Nvidia moat is CUDA sitting between PyTorch and the silicon, and that this is problematic over time. The fix he is building at 8090 is a functional transpiler that recompiles CUDA-written code to any accelerator without losing performance, which would make AI workloads portable to Amazon's own silicon and Google TPUs and strip the software lock-in out of Nvidia's pricing power.
0
CONVICTION
peaked 31.6 EMBER
DORMANT
band · contested
▼ SHORT NVDA
expression · bearish tech
MISS
outcome · R -62.5% · α -83.8%
2025-09-06
window closed
⚠ CONFLICTING IDEAS ON THE SAME TICKER
— the board is arguing with itself; net it before trading (ticker view →)
NVDA ·
AI compute capex supercycle has 12-24 months of runway
LONG
GREEN 86.6
← stronger than this idea (2.1)
NVDA ·
Nvidia + Hugging Face makes NVDA the open-source AI champion
LONG
WATCH 57.9
← stronger than this idea (2.1)
GOOGL ·
Anthropic IPO (~Oct 2026) is attractively priced at $1.5-2T
LONG
WATCH 53.9
← stronger than this idea (2.1)
GOOGL ·
Personal AI agents: Google's data moat wins the form factor
LONG
EMBER 29.3
← stronger than this idea (2.1)
GOOGL ·
Frontier AI labs keep durable premium pricing (not commoditized)
LONG
EMBER 26.6
← stronger than this idea (2.1)
GOOGL ·
Own AWS, Azure and GCP and nothing else — the clouds capture the AI application dollars
LONG
EMBER 20.6
← stronger than this idea (2.1)
AMZN ·
Own AWS, Azure and GCP and nothing else — the clouds capture the AI application dollars
LONG
EMBER 20.6
← stronger than this idea (2.1)
GOOGL ·
Independent frontier-model labs lose to hyperscaler capital
LONG
DORMANT 14.0
← stronger than this idea (2.1)
AMZN ·
Independent frontier-model labs lose to hyperscaler capital
LONG
DORMANT 14.0
← stronger than this idea (2.1)
GOOGL ·
Willow's error-correction scaling puts encryption on a two-to-five-year clock
LONG
DORMANT 8.8
← stronger than this idea (2.1)
GOOGL ·
Meta's AI assistant takes ten points of search share from Google
LONG
DORMANT 7.4
← stronger than this idea (2.1)
also touching these tickers, same direction: Blue cities ban or license-cap robotaxis 28 (GOOGL) · Tort floodgates open on social media 15 (GOOGL) · Nvidia risk: special-purpose silicon + Huawei 10 (NVDA) · ChatGPT takes Google search share and the market starts pricing the decay 9 (GOOGL)
⚖ Why this verdict
fully deterministic — evaluate.py replays this from daily closes; nothing below is editable or hand-set
- Window: 2024-09-06 → 2025-09-06 — first mention + 12-month horizon, then the window locks.
- The call: ▼ SHORT NVDA (primary play). NVDA rose 62.5% over the window → direction-adjusted R = -62.5% (the call lost money).
- Benchmark: holding SPY over the same window returned +21.3%
→ α = -62.5 − (+21.3) = -83.8%
— what this call made or lost against just owning the index. This is the number the verdict uses.
Stock-selection read: -41.2% — did NVDA move the predicted way relative to the market. The two only differ on a SHORT: this call was right about direction relative to the index but still cost money versus holding it.
- Rule fired:
· HIT — R ≥ +10% AND α ≥ +5· PARTIAL — R ≥ +5% OR α ≥ 0▶ MISS — everything else
- Credit: supporters of a MISS earn 0 each, opposers the inverse — this feeds the scoreboard weights. supported: nobody | opposed: Chamath
Conviction timeline
bands: green ≥ 65 · watch ≥ 45 · ember ≥ 15
Plays vs SPY · % since first mention (2024-09-06)
Plays
| expression | symbol | kind | relevance | rationale |
|---|---|---|---|---|
| ▼ SHORT | NVDA | stock | PRIMARY | the CUDA owner whose software lock-in the transpiler is designed to remove |
| ▼ SHORT | AMD | stock | adjacent | the merchant alternative that only wins once CUDA code compiles elsewhere |
| ▼ SHORT | AMZN | stock | adjacent | Amazon's own accelerators are the other named beneficiary |
| ▼ SHORT | GOOGL | stock | adjacent | TPU is one of the two named targets that gains if code becomes portable |
Mention log
Chamath
tech w=0.90 · n=83
· E232 (2025-06-21) · positioning · strength 3
▶ 22:14
REVERSAL
“When I first started 80-90 a year ago, one of the key bets I made, which was a mistake and we unwound the bet, but the first bet that I made was, can we build a transpiler? ... And basically what I learned in that process are all of the attention mechanisms that are built into transformers that really differentiate how good the models are, need to literally be hand-tuned for every single target of silicon that you have.”
Who built this conviction
each voice's total force on the score — supports and opposes from every mention, weighted exactly as the replay applied them · share = % of all mention-driven movement
⏳ decay drained -25.0 over the idea's life — that's time passing, attributed to no one
Score events
| episode | kind | Δ | after | note |
|---|---|---|---|---|
| E195 2024-09-06 | init | +31.6 | 31.6 | E195 born by Chamath (explicit_prediction x2) [w=0.90] |
| E196 2024-09-20 | decay | -1.3 | 30.4 | E196 silent |
| E197 2024-09-27 | decay | -1.2 | 29.1 | E197 silent |
| E198 2024-10-03 | decay | -1.2 | 28.0 | E198 silent |
| E199 2024-10-11 | decay | -1.1 | 26.9 | E199 silent |
| E200 2024-10-18 | decay | -1.1 | 25.8 | E200 silent |
| E201 2024-10-25 | decay | -1.0 | 24.8 | E201 silent |
| E202 2024-11-01 | decay | -1.0 | 23.8 | E202 silent |
| E203 2024-11-08 | decay | -1.0 | 22.8 | E203 silent |
| E204 2024-11-16 | decay | -0.9 | 21.9 | E204 silent |
| E205 2024-11-23 | decay | -0.9 | 21.0 | E205 silent |
| E206 2024-12-07 | decay | -0.8 | 20.2 | E206 silent |
| E207 2024-12-13 | decay | -0.8 | 19.4 | E207 silent |
| E208 2024-12-20 | decay | -0.8 | 18.6 | E208 silent |
| E209 2025-01-04 | decay | -0.7 | 17.9 | E209 silent |
| E210 2025-01-11 | decay | -0.7 | 17.1 | E210 silent |
| E211 2025-01-18 | decay | -0.7 | 16.5 | E211 silent |
| E212 2025-01-25 | decay | -0.7 | 15.8 | E212 silent |
| E213 2025-01-31 | decay | -0.6 | 15.2 | E213 silent |
| E214 2025-02-07 | decay | -0.6 | 14.6 | E214 silent |
| E215 2025-02-15 | decay | -0.6 | 14.0 | E215 silent |
| E216 2025-02-21 | decay | -0.6 | 13.4 | E216 silent |
| E217 2025-03-01 | decay | -0.5 | 12.9 | E217 silent |
| E218 2025-03-08 | decay | -0.5 | 12.4 | E218 silent |
| E219 2025-03-15 | decay | -0.5 | 11.9 | E219 silent |
| E220 2025-03-22 | decay | -0.5 | 11.4 | E220 silent |
| E221 2025-03-29 | decay | -0.5 | 10.9 | E221 silent |
| E222 2025-04-05 | decay | -0.4 | 10.5 | E222 silent |
| E223 2025-04-11 | decay | -0.4 | 10.1 | E223 silent |
| E224 2025-04-19 | decay | -0.4 | 9.7 | E224 silent |
| E225 2025-04-26 | decay | -0.4 | 9.3 | E225 silent |
| E226 2025-05-02 | decay | -0.4 | 8.9 | E226 silent |
| E227 2025-05-09 | decay | -0.4 | 8.6 | E227 silent |
| E228 2025-05-17 | decay | -0.3 | 8.2 | E228 silent |
| E229 2025-05-24 | decay | -0.3 | 7.9 | E229 silent |
| E231 2025-06-13 | decay | -0.3 | 7.6 | E231 silent |
| E232 2025-06-21 | reversal | -4.5 | 3.0 | E232 Chamath flips |
| E233 2025-06-28 | decay | -0.1 | 2.9 | E233 silent |
| E234 2025-07-04 | decay | -0.1 | 2.8 | E234 silent |
| E235 2025-07-11 | decay | -0.1 | 2.7 | E235 silent |
| E236 2025-07-19 | decay | -0.1 | 2.6 | E236 silent |
| E237 2025-08-01 | decay | -0.1 | 2.5 | E237 silent |
| E238 2025-08-09 | decay | -0.1 | 2.4 | E238 silent |
| E239 2025-08-15 | decay | -0.1 | 2.3 | E239 silent |
| E240 2025-08-22 | decay | -0.1 | 2.2 | E240 silent |
| E241 2025-08-29 | decay | -0.1 | 2.1 | E241 silent |