Tesla's forthcoming Dojo supercomputer could adhd arsenic overmuch arsenic $500 cardinal to nan automaker's valuation – which presently sits astatine $875 cardinal - according to a Morgan Stanley Research note.
The financial services elephantine is bullish connected Dojo arsenic it feels Tesla's improvement of civilization silicon for nan ace will person applications good beyond improvement of full-self driving (FSD) capabilities.
What do you mean your exaflop is amended than mine?
"Investors person agelong debated whether Tesla is an auto-company aliases tech company. We judge it's both, but spot nan biggest worth driver from present being package and services revenue," nan Morgan Stanley statement stated. "If Dojo tin thief make cars 'see' and 'react,' what different markets could unfastened up? Think of immoderate instrumentality astatine nan separator pinch a camera that makes real-time decisions based connected its ocular field." And insert Dojo, presumably.
Morgan Stanley has truthful accrued its target value for Tesla shares from $250 to $400 apiece.
Developed in-house by Tesla, astatine a costs of possibly $1 cardinal aliases more, Dojo's architecture bears small resemblance to wide intent AI accelerators designed by nan likes of Nvidia, AMD, and Intel. Instead, nan full strategy — compute, networking, IO, moreover nan instruction group — were developed to process ample quantities of telemetry gathered by nan galore sensors and camera built into Tesla vehicles.
The supercomputer was said to beryllium put into accumulation complete nan summer.
Dojo’s cardinal constituent is nan D1 AI accelerator processor, which features 354 civilization CPU cores. Twentyfive of these chips are meshed successful a 5x5 grid utilizing TSMC's chip-on-wafer-on-substrate (CoWoS) packaging tech to shape nan Dojo training tile. As Tesla explained astatine 2022’s Hot Chips conference, six Dojo training tiles are networked together to shape nan V1 system, which is supposedly tin of 1 exaFLOPS of BF16 compute.
Tesla CEO Elon Musk claimed successful July this twelvemonth that nan complete Dojo supercomputing cluster could transcend 100 exaFLOPS erstwhile built.
- GenAI proponents exploring its usage alongside infrastructure-as-code
- Google thinks $20M ought to beryllium capable to fig retired really aliases if AI tin beryllium utilized responsibly
- Microsoft's AI investments skyrocketed successful 2022 – and truthful did its h2o consumption
- Power grids tremble arsenic electrical conveyance maturation group to accelerate 19% adjacent year
"With importantly accrued computing powerfulness and faster processing speeds (latency) Tesla's way to monetizing conveyance package tin materialize sooner," nan Morgan Stanley investigation statement continued. "We expect caller charging position practice will widen into FSD licensing (discussions ongoing) and operating strategy licensing."
In different words successful summation to being capable to waste Tesla vehicles, Dojo could spot Tesla go an AI-as-a-service supplier to automakers that request FSD capabilities.
Before it gets into nan SaaS business Tesla needs to decorativeness building Dojo.
While it useful connected nan super, Tesla has invested successful replacement AI infrastructure. Last month, nan automaker revealed a cluster packing 10,000 of Nvidia's fastest and astir costly H100 accelerators. The deployment is astir doubly nan size of nan automaker's erstwhile A100-based compute cluster and should present astir 40 exaFLOPS of highest FP8 performance. ®