How AI costs impact media workflows

The latest Pulsecast episode, hosted by Lorenzo Zanni of Dovetail Creative, sheds light on a question many media professionals are asking: how much does artificial intelligence (AI) really cost? In Episode 10, “The AI Pricing and Cost,” Damon Neale, CEO of Metadat-AI, explains the ins and outs of AI pricing models and the factors behind their steep costs, particularly for media and entertainment (M&E). Released in May 2025, this episode is essential listening for anyone wondering how to budget for AI.

At Metadat-AI, we focus on making AI practical and accessible for M&E. This article highlights the episode’s main takeaways, showing how Damon’s insights can help media businesses make informed AI investments.

Why AI Costs More Than You Think

AI isn’t cheap, and Damon explains why with clarity. The biggest cost driver is graphics cards, or GPUs, like the Nvidia A100, which cost around $10,000 each. Unlike standard CPUs, which are about $1,000, GPUs are essential for training and running AI models. The demand from companies like OpenAI and even governments building $20 billion “gigafactories” full of GPUs has created a supply shortage, pushing prices up. These cards also use a lot of energy and need expensive cooling systems, adding to the bill.

Then there’s the talent. AI experts are rare, with salaries ranging from $120,000 to over $300,000 a year. On top of that, building an AI product involves the same costs as any software-as-a-service (SaaS) tool—think support staff, sales, and management—plus the extra expense of AI research and data storage. For M&E, where processing video or managing assets is compute-heavy, these costs hit hard.

How AI Pricing Works: Models Explained

Damon simplifies AI pricing models, which can feel like a maze. Unlike traditional SaaS, which often charges a flat $20 per month per user—like ChatGPT’s subscription—AI pricing is more varied. The episode highlights several models, with transactional pricing being the most common in M&E. This means you pay per unit of content processed, like $0.01 per image or $10–$25 per hour of video. For tasks like analyzing sports footage or subtitling, these costs add up fast.

Another model is credit-based, where you buy credits upfront, say $100 for 100,000 credits, and each task consumes them differently. Optical character recognition (OCR) might be cheap, but describing video scenes is pricier. Damon notes this can confuse clients, as it’s hard to predict costs. Some vendors charge based on storage, like per gigabyte in a media asset management (MAM) system, bundling AI processing for “free.” There’s also user-based pricing, like ChatGPT’s $20/month, but with limits on features like image generation. Lastly, renting cloud GPUs for open-source models can cost $45/hour for high-end setups, though Damon’s team at Metadat-AI uses machines as low as 50 cents/hour for comparable results.

The Polarization of AI Pricing

Lorenzo and Damon discuss a clear trend: AI pricing is splitting into two worlds. Basic models, like ChatGPT, are affordable at $20/month, almost a commodity for consumers. But specialized tools for businesses, especially in M&E, are pricier—think $17/hour for top-tier video processing. Damon sees this polarization growing. Consumer prices may stay low to gain market share but business prices could rise, reflecting AI’s value in replacing costly human roles.

Open-source models, like DeepSeek, shake things up. When DeepSeek released a free ChatGPT rival, it cut Nvidia’s value by 17% overnight, showing how open-source can keep prices in check. But Damon points out that productizing these models—adding support, hosting, and updates—still costs money, limiting how low prices can go. For M&E, where live video or compliance tasks demand precision, premium models are often worth the cost, but open-source options can work for simpler needs.

What’s Next for AI Pricing?

Looking ahead, Damon predicts changes over the next three to seven years. Transactional costs, like $17/hour for video processing, should drop as GPU supply improves and government-backed gigafactories reduce infrastructure costs. Open-source models and efficient hardware, like specialized chips for specific tasks, will help. Damon’s team already runs AI on 50-cent/hour machines, a game-changer for M&E’s price-sensitive workflows.

However, businesses might face higher costs for specialized AI, especially in high-stakes fields like legal or medical, where accuracy is critical. Damon warns of an “Uber effect,” where low prices—like ChatGPT’s $20/month—rise after companies dominate the market. For businesses, AI licenses could cost $1,000/month to replace a $36,000/year employee, a bargain for some but a shift from today’s pricing. Meanwhile, more AI experts entering the field, thanks to new degrees and prompt engineering skills, could lower talent costs, benefiting M&E budgets.

Why Infrastructure Matters

The episode also touches on AI’s infrastructure boom. Companies and governments are pouring billions—$20 billion in the EU, $52 billion in China—into data centers packed with GPUs. These “gigafactories” lock in high costs for the next three to five years, as equipment is written off over that period. Energy and cooling expenses keep rising, but Damon sees hope in alternatives like his low-cost machines or specialized chips, like Tesla’s for self-driving cars, which could cut costs for media tasks.

Listen and Learn with Metadat-AI

Pulsecast Episode 10 is a goldmine for M&E professionals addressing AI’s costs. Damon Neale’s insights make sense of pricing models, from transactional fees to open-source savings, and offer a glimpse into the future. Whether you’re a producer, editor, or manager, this episode shows how to budget for AI without losing sight of quality.

At Metadat-AI, we believe AI should be practical and accessible. Our Introduction to Artificial Intelligence course builds on Damon’s expertise, teaching you to choose cost-effective AI tools for media tasks like compliance checks or asset management. With our in-cloud platform, you can start using AI today, keeping costs low and data secure.

Listen to Pulsecast Episode 10 on https://lnkd.in/dbWquCJw and visit Metadat-AI.com to learn how our course can help you master AI for your media projects.