Tesla Hits an AI Silicon Milestone: Musk Says AI5 Has Taped Out; Dual-Chip Performance Said to Rival Blackwell

Tesla’s in-house silicon push reportedly reached a major step: Elon Musk said the next-generation AI5 has taped out, with high-volume manufacturing targeted around 2027 at Samsung and TSMC fabs in the United States. The same narrative frames AI5 as up to ~40× better than AI4 on headline metrics, with a dual-chip configuration described as competitive with NVIDIA Blackwell on throughput while claiming better cost and power.

Tape-out announced; dual-chip positioning versus Blackwell

Tesla describes the moment as a key milestone for its vertically integrated AI silicon. On X, Musk said AI5 has taped out, that the design handoff to foundries is underway, and that production is expected to start in 2027. AI5 is expected to succeed AI4 as the primary compute platform for Full Self-Driving (FSD) and humanoid robotics workloads.

Musk also said AI5 would be built by Samsung and TSMC in U.S.-based manufacturing footprints. In the same thread, he reportedly mistagged TSMC’s account, briefly pointing to a similarly named account and causing short-lived confusion on social.

Citing prior reporting attributed to TradingKey, coverage claimed AI5 single-chip performance in the ballpark of NVIDIA Hopper, while a dual-chip setup is described as approaching Blackwell-class levels, with lower cost and power than comparable NVIDIA parts. Musk has also claimed AI5 could improve key metrics by ~40× versus AI4, including on the order of 9× memory and 8× compute (figures attributed to Musk in third-party summaries).

Musk further indicated that AI6 and Dojo 3 remain on the roadmap, portraying Tesla’s AI silicon roadmap as continuing in parallel tracks.

Tape-out complete; production targeted for 2027

Tape-out is the late stage of chip development where the design is submitted for manufacturing. If a 2027 start-of-production target holds, that implies roughly ~12 months from early 2026 commentary to first meaningful volume—subject to yield, packaging, software bring-up, and supply-chain realities.

AI5 is positioned as AI4’s successor: the compute backbone for autonomy and robotics programs. Reporting attributed to TradingKey also framed AI5 as a strong fit for inference on models under ~250B parameters. Separately, Tesla is discussed as continuing broader AI infrastructure bets—including work with Intel around Terafab-style capacity narratives.

Reporting attributed to TechPowerUp described manufacturing split across Samsung (notably Taylor, Texas) and TSMC (Arizona), emphasizing domestic wafer capacity as a theme.

Versus AI4: claimed ~40× lift; dual-chip versus Hopper/Blackwell framing

On performance, summaries citing TradingKey describe single-chip AI5 as Hopper-class and dual-chip as approaching Blackwell-class, again paired with claims of better cost and power.

Musk has been quoted framing AI5 as existential for Tesla—requiring focused execution across teams, including weekend work—while also claiming room to re-accelerate Dojo 3 once AI5 risk is reduced:

“AI5 is going well enough that we finally have some bandwidth to restart serious Dojo 3 work.”

Risk disclosure

Markets involve risk; this is not individualized investment advice. Summaries may omit context, caveats, and forward-looking uncertainties. Verify claims against primary sources (e.g., official Tesla communications and filings) before making decisions.