The Power Grid’s AI Problem, Plus Meta and Google take on ChatGPT 4/18/24

TechCheck

CNBC

Management, Cnbc, Tech, Faang, Investing, Business, Disruptors, Technology

4.5 • 66 Ratings

🗓️ 18 April 2024

⏱️ 7 minutes

🔗️ Recording | iTunes | RSS

🧾️ Download transcript

Summary

: The latest bottleneck in the AI boom might not be chips anymore - it's energy. Computing for AI requires way more power than traditional data centers, and CEOs including Elon Musk, Andy Jassy and Sam Altman are all sounding the alarm. Plus, the AI race is on -- and it's now faster and more competitive than ever. Meta says its just-released latest AI model, the Llama 3, is the most powerful model that's also freely accessible. Meta is also teaming up with Google to include search results with its new AI chatbot, Meta AI, in a show of force against ChatGPT.

Transcript

Click on a timestamp to play from that location

0:00.0	T.SMC now warning that rising electricity costs in Taiwan are impacting profitability and
0:06.4	power grid capacity could also be the latest supply constraint for the AI boom in America too.
0:12.0	Our Christina Parts of Nevilo is taking a look at that for
0:13.9	today's tech check. Hey K. P. Hi Carl. Well you said
0:17.1	a T.S. M. did beat revenue and profit expectations in its first quarter but
0:20.7	soaring electricity costs are eating into profits.
0:24.7	The company forecasts a 25% increase in electricity costs starting this past April that would
0:29.2	reduce gross margins by up to 0.8 percentage points and also contribute to higher costs for
0:34.8	customers and competitors according to the CEO and the earnings call this morning.
0:38.6	T.S.M.C's power costs are emblematic though of a greater problem facing firms all around the United States.
0:45.0	How are they going to power these AI machines when so much of the infrastructure is outdated?
0:51.4	Several analysts I spoke to at NVIDIA's recent events called it the
0:54.3	power gate and the power gate was going to be the biggest cost hurdle going forward.
0:58.4	AI, we know is driving the construction of large computing warehouses that require way more power than
1:04.2	traditional data centers.
1:05.6	McKinsey, for example, says that data center electricity use will double over this decade
1:10.1	to 8% of all US power demand, up from 4%.
1:14.0	The latest International Energy Agency report stated that
1:18.0	CHATGPT consumes roughly 2.9 watts per hour of electricity per request.
1:23.5	What that really means is running a 60 watt light bulb for three minutes for one
1:28.3	chat tPT request.
1:30.0	That's 10 times as much electricity as a Google search.
	...

Please login to see the full transcript.

Previous episode | Next episode

Disclaimer: The podcast and artwork embedded on this page are from CNBC, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of CNBC and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.