• The AI Plug
  • Posts
  • # 7 ChatGPT operational costs at $700,000 per day

# 7 ChatGPT operational costs at $700,000 per day

AND: No AI chips for China

10 years ago, Jensen Huang didn’t know what would happen with computing power - but he and Nvidia’s board of advisors knew something would happen.

That premonition is why they went all-in on accelerated computing - designing components which made computers process faster.

Now, for better or worse - they provide the horsepower which is driving the AI revolution.

In today’s newsletter:

  • ChatGPT operational costs at $700,000 per day: Nvidia reaping the benefits of early investment in AI

  • The rundown: K-12 curriculum updates, AI bias and China struggling to keep up with global AI race

  • Tips, tricks & tools: Create your own commercials, and generate videos of yourself from your photographs

  • No AI chips for China: U.S. has banned exports of leading-edge AI chips to China

  • Perhaps the sun will save us? A comic reflection on what MIGHT be in store 

CHATGPT OPERATIONAL COSTS AT $700,000 PER DAY

Nvidia provides the processing chips which power ChatGPT servers, each of which cost $10,000

We saw early on, about a decade or so ago, that this way of doing software could change everything. And we changed the company from the bottom all the way to the top and sideways. Every chip that we made was focused on artificial intelligence.”

Jensen Huang, CEO of Nvidia, the world’s 7th most valuable company with a market cap of $555 billion, is best known for designing and manufacturing graphics processing units (GPUs).

What might surprise you, is that Nvidia is also basically powering the AI revolution with their computer chips.

Just one of Nvidia’s top chips, the A100, comes in at around $10k. It’s these chips that enable computers to process the data necessary to run these AI models, and with AI as hot as it is, computational capacity is the most sought-after currency in the industry.

The computing power required by OpenAI’s ChatGPT, for example, costs an estimated $695,000 per day to operate in computing hardware costs, using 28,936 GPUs.

Limitations in computing power is also what’s currently holding AI researchers back from further advances in the field - but Huang and Nvidia already have a solution to that - the H100.

Weighing in at 50 pounds, the Hopper 100 is a system designed with transformer engine processing and is, according to Huang, “the world’s first computer designed to process transformers at enormous scale. So large language models are going to be much, much faster and much more cost-effective.”

If we think we’ve seen what AI can do - we’re very, very wrong.

With the funding, the hardware and the knowledge which currently exists around AI, the next few months, let alone years, are going to see some truly awe-inspiring developments.

Awesome? Yes. 

Terrifying? Absolutely.

THE RUNDOWN 🐂

  • One-size-fits-all K-12 curriculums have been an issue with education, something which AI can address - helping students get better feedback, particularly on reading and writing skills. (link)

  • SlackGPT - As Salesforce announces a partnership with ChatGPT’s OpenAI, consumers can expect to see more AI chatbots in SaaS and other platforms like Slack. (link)

  • Biases in AI highlighted by AI photo studio. (link)

  • Unlock powerful insights from your own image and video content using Microsofts Azure Cognitive Service for Vision. (link)

  • Talk face-to-face with photorealistic AI with D-ID’s web app. (link)

  • AI can mitigate the effects of climate change. (link)

  • Grocery and other perishable goods vendors to use ChatGpt. (link)

TIPS, TRICKS & TOOLS ⚒️

  • D-ID - Users can talk face-to-face with the Israeli startup’s AI chatbot, thanks to a mix of video-to-streaming and ChatGPT technology. (link)

  • Deep Agency - Hire and create virtual models for virtual photo services with advanced AI technology. (link)

  • Waymark - easily create unique, high-converting AI-powered commercials for your business. Proven to boost local sales. (link)

  • Gen AI videos - Create unique videos of yourself using your still images. (link)

  • Quotify - Find and extract meaningful quotes on any topic from any book, article, or text in this database. (link)

NO AI CHIPS FOR CHINA

As tensions between the U.S. and China rise (particularly in the disputed South China Sea, where jets are shadowing one another and instructing pilots to remove themselves from airspace), we can see why the U.S. decided to restrict the export of certain semiconductors and computational processing units to China last year.

Introduced in October 2022, the regulations imposed by the U.S. mandated that:

  • Companies, such as Nvidia, need special licenses to export high-performance chips (like those used to power AI systems) to China

  • The exportation of machinery to Chinese chip-processing companies is heavily restricted.

These restrictions are designed to hamper the Chinese manufacturing and development of computer chips essential for supercomputers - something which seems to have effectively slowed the country’s ability to develop AI systems capable of performing to the level of ChatGPT.

In particular, the ability of AI systems to deliver results in real-time is “very difficult to achieve,” according to China’s Science and Technology Minister Wang Zhigang.

As discussed earlier in this newsletter, the computational and hardware demands for running such AI systems are enormous.

With U.S.-based company Nvidia currently providing 95% of the processing chips used in AI systems, the restrictions upon U.S. companies mean China can’t get the same tools as everyone else - at least not for now.

THERE’S A CHANCE THIS ISN’T TOO FAR FETCHED

The sheer electromagnetic power of solar flares can disrupt power grids, internet connections and other earthly-communication devices, so a big one could impact AI systems.

Will anything like this play out though? 

SPREAD THE WORD

Guys, we’ve got a favour to ask.

We’re a new newsletter, and we want to keep the lights on.

As a valued subscriber to The AI Plug, it would really help if you’d share this newsletter with a friend (or friends).

All you have to do is send you’re unique Ambassador URL (⏬) to anyone you think might enjoy the newsletter as well!

What’s more, if 4 people subscribe using your link, YOU gain access to our exclusive online community, where members get to:

  • Network with like-minded individuals

  • Chat directly with us about any AI ideas/thoughts you have

  • Gain early access and discounted subscriptions to the latest AI tools

FEEDBACK