Thinking through AI as the next new platform opportunity - Version One: Why Databricks Bought Mosaic and The Rapidly Decreasing Costs of AI

Thinking through AI as the next new platform opportunity - Version One: Why Databricks Bought Mosaic and The Rapidly Decreasing Costs of AI

Artificial intelligence (AI) has become a hot topic in recent years, with many companies recognizing its potential to revolutionize industries and create new opportunities. As AI continues to evolve, there are several important questions to consider.

One of the most crucial questions is how much value the underlying AI platform will keep for itself. History has shown that most platforms extract the bulk of the value, leaving only crumbs for the companies and developers building on top. However, in the case of AI, there is a good chance that several companies will offer similar platforms, similar to how cloud computing is divided between AWS, Azure, and GCP. OpenAI, Google, and Facebook are leading contenders to play significant roles in the AI platform space.

Competition between these companies should limit a single company's potential to maximize profits at the expense of companies building on top. This is an encouraging sign for developers and companies who want to leverage AI technologies without being overly reliant on a single platform.

The second question to consider is who will capture most of the new value being generated: incumbents or start-ups? Incumbents have the advantage of leveraging their unique data sets, large user bases, and balance sheets to gain an edge, especially on the consumer side. However, when it comes to enterprise software, there is an opening for start-ups to create more value for their customers with vertical-specific AI products.

Start-ups can differentiate themselves by designing interaction models that minimize friction and risk, such as making it easy to bring a human into the loop in cases where false positives or false negatives may occur. Additionally, start-ups can insert proprietary data into the AI models, offering unique insights and capabilities. Differentiating with original datasets and data from feedback loops is frequently mentioned as a key strategy. Addressing regulatory and privacy constraints is also crucial for companies to protect their brands.

In terms of the cost of AI, there have been significant advancements that make it more accessible for companies to train and fine-tune their own models. Databricks' acquisition of Mosaic aligns with their vision to help companies rapidly adopt machine learning and outpace the competition. Mosaic's vision to make it cost-effective for companies to train their own models is accelerating, thanks to general trends in decreasing hardware costs.

Training costs have decreased significantly, with stable diffusion training costing $50k compared to $600k in the past. This decrease in costs can be attributed to algorithmic improvements made by companies like MosaicML and the decreasing costs of GPUs, which have gone down 3x in about 3 years. The decreasing cost of inference is also worth noting, with a 10x decrease in 16 months.

These decreasing costs have significant implications for companies in terms of their cloud expenses. Most SaaS companies target 80% gross margins, with cloud costs typically accounting for around 10% of revenue. However, AI features are in addition to regular cloud costs, so companies may end up spending a similar amount on cloud costs as they do on generative AI features.

The decreasing costs of training models and the rise of open-source model providers will lead to more competition at the model layer. This should place pricing pressure on closed-source model providers and encourage more companies to start with open-source options.

In conclusion, AI presents a new platform opportunity with the potential for significant value creation. The competition between leading companies like OpenAI, Google, and Facebook should prevent any one company from monopolizing the space. Start-ups also have the chance to create value by leveraging vertical-specific AI products and addressing regulatory and privacy constraints.

To navigate the AI landscape effectively, here are three actionable pieces of advice:

1. Diversify your AI platform partnerships: Instead of relying on a single platform, consider partnering with multiple AI platforms to mitigate any potential risks and take advantage of different capabilities.

2. Focus on vertical-specific AI products: If you're a start-up, consider developing AI products that cater to specific industries or niches. This can give you a competitive edge and create more value for your customers.

3. Stay informed about cost-saving opportunities: Keep track of the decreasing costs of AI training and inference, as well as the rise of open-source model providers. This knowledge can help you make informed decisions about your AI strategy and potentially save costs.

By considering these factors and taking action accordingly, companies and developers can position themselves for success in the rapidly evolving AI landscape.