The tech market and the investor community are foaming at the mouth about the impact and implications of AI. Every podcast, white paper, blog, and forum is awash with conflicting opinions on how AI is either going to destroy the tech industry or expand the markets infinitely. It is either the end of tech investing or the beginning of a new era. It will either eliminate all jobs, or it will make all jobs better. This is the SaaS-pocolypse and the death of all SaaS companies, or it is the rebirth of SaaS companies. The bottom line is that we are currently in a state of AI confusion and chaos.
While it is certainly a challenging time for business leaders and investors alike, my favorite concept is that all creative solutions come at the interface of chaos and order. Albert Einstein, Steve Jobs, and Nikola Tesla all emphasized that creativity involves making order out of chaos, but Mary Shelly (of Frankenstein fame) probably said it best when she noted that “invention does not consist in creating out of void, but out of chaos,” suggesting that the creative process involves seizing the potential within disorder. Chaos is where all the magic happens. Out of the AI chaos we are starting to see new areas of order emerge and take shape. I recently attended a CFO meeting where the discussion centered on how to properly code and account for AI costs both in costs of goods sold and in operations. As accountants standardize the mechanisms for tracking AI costs, the analysts are forming new frameworks for calculating performance metrics. As an example, one model suggests we calculate dollars of revenue per dollar of AI spend as a ratio that should be increasing in a business that is scaling and becoming more AI efficient. Historically, human resources have been the largest cost element in software companies, followed by systems and computing costs. We now have rapidly rising AI costs that in some companies have eclipsed the baseline computing costs. Hence, as we look at revenue per employee, we should also consider revenue per AI cost.
My favorite topic has been the variability of AI costs and how this adversely impacts gross margin, particularly if it is being deployed as a component of a customer facing product. Companies that are heavily AI-forward in their product offerings are currently experiencing dramatic incremental costs of goods sold, which is leading to reduced gross margins. By some estimates, a 20% reduction in gross margin, or more. One of the accounting suggestions is to track the AI contribution to COGS and separately report baseline gross margin excluding AI costs and net gross margin including AI costs. This provides a view of base unit economics and enables analysis of the impact AI is making on the business.
With lower gross margins, there is less money to cover operating expenses and still maintain profitability. Savvy companies are focusing on productivity gains, particularly in engineering costs to build and support their products, as their way to absorb the decreased gross margin impact. Some analysts are betting on AI costs decreasing as the industry scales and becomes more competitive. In my humble opinion, I think that is a long way off. We are seeing AI companies make massive investments in infrastructure and research, and they are scaling their pricing to pass along costs by either charging a higher fee for tokens, or changing the equation so that solving the same problem consumes more token to complete. At the same time, application vendors are expanding the role of AI in their products, which is leading to increased demand for tokens, so even if the cost of tokens declines, vendors will still experience rising AI costs based on consumption. The structural change I see on the horizon is a rapid move toward consumption-based pricing for customers of the application vendors. Vendors cannot absorb the expanding costs of AI that they are delivering to their customers, so the only answer is for customers to bear the cost. CFOs and procurement managers are loath to accept variable consumption pricing, but it seems inevitable that these attitudes will have to change.
We are already seeing this shift as several giant SaaS vendors are renewing licenses with built-in token costs and escalators when a customer exceeds their included tokens. Vendors are delivering these new licenses with estimates of what a customer’s consumption will be, but most customers are already realizing that they will greatly exceed the estimates provided.
The bottom line is that once we know how to instrument the new forms of AI-enabled businesses, we can learn how to creatively drive order from the current chaos. Vendors are figuring out how to manage their profitability in the age of AI, and they are also figuring out how to institute changes in the way they price and grow revenues. We are not out of the chaos of AI, but we are approaching the boundary between chaos and order where all innovations emerge.