AMD just made bold move to challenge Nvidia

Advanced Micro Devices (AMD) has spent 2026 proving it tin vie with Nvidia successful the artificial quality race. 

Its banal has more than doubled this year, and its information halfway concern is increasing fast.

Now AMD is trying thing different.

The institution conscionable bought a tiny Toronto startup with a extremist thought astir however AI chips should beryllium built. 

The woody volition not determination AMD's gross this quarter, and it volition not dent Nvidia's pb close away.

But it points to wherever the adjacent signifier of the AI buildout is heading, and it could substance a large woody for the unreality companies that are moving retired of inexpensive electricity.

Here is what AMD bought, wherefore it did it, and what it means if you ain the stock.

AMD acquires Taalas to onslaught Nvidia's grip connected AI chips

Advanced Micro Devices announced connected August 6 that it agreed to get Taalas, a Toronto startup that builds chips for AI inference. Financial presumption were not disclosed.

Inference is the enactment of moving a trained AI exemplary to reply questions, the portion users really interaction erstwhile they benignant a prompt.

More AI Chip Stocks:

• AMD banal falls aft grounds 4th arsenic SpaceX picks Nvidia

• Nvidia banal is doing thing it hasn't done successful years

• Bank of America sees Nvidia's adjacent $20 cardinal business

Nvidia (NVDA) controls roughly 80% to 90% of the information halfway AI spot market, and its CUDA bundle keeps developers locked into its hardware.

By buying Taalas, AMD is signaling that general-purpose graphics processing units, the flexible chips that Nvidia dominates, are nary longer the lone mode to triumph successful AI.

Nvidia paid about $20 billion for assets from inference startup Groq astir seven months earlier, CNBC reported. 

Both companies are present buying their mode into the aforesaid fast-growing country of the market.

AMD's acquisition of Taalas deepens its propulsion into the AI inference market, wherever it is trying to adjacent the spread with Nvidia.JHVEPhoto / Getty Images

What Taalas really built

Most AI chips, including Nvidia's, are general-purpose. 

They tin tally immoderate model, but they perpetually shuttle billions of exemplary weights betwixt the processor and costly memory, which wastes clip and power.

Taalas takes the other approach. It etches a azygous AI model's weights straight into the silicon.

Its archetypal chip, the HC1, runs Meta's (META) Llama 3.1 model and thing else, The Register reported. 

Related: SoftBank CEO offers stinging critique of Musk's AI bet

Built connected TSMC's (TSM) 6-nanometer process, Taalas says it tin make more tokens per 2nd than Nvidia's H200 and B200 chips portion utilizing one-tenth of the power, DatacenterDynamics reported.

Here is the catch. A Taalas spot is limited to 1 model. Switch models, and you request caller silicon.