The AI "Race to the Bottom"
What happens when Big Tech gives away chatbots for free?
In yesterday’s Microsoft Build Keynote, one announcement caught my eye in particular - that Microsoft is giving away (for free) a GPT-4 powered AI tutor application called Khanmigo to every K-12 teacher in the U.S.
Essentially, Khanmigo is an AI tool for teachers to create lesson plans, etc, and has been around for a year. What changed is that Microsoft will now offer it for free. This “AI tools for teachers” space was already crowded with multiple venture backed apps like Magicschool.ai, etc. So it remains to be seen how this space consolidates.
So why was this move of giving away Khanmigo interesting? Because it signals that the price of GenAI apps may be on a “race to the bottom”, aka $0. There are legions of AI startups founded on the assume that better UX and data can justify selling a $20/mo SaaS. This will be challenged if Microsoft or Google starts offering free GenAI tools for specific verticals.
Microsoft isn’t alone in this “free GenAI app trend” - OpenAI will offer GPT-4o powered ChatGPT for free, and Google will also give away practically every GenAI launch from Google I/O.
In this post, I’ll talk about:
whether this Microsoft move is just a one-off event, or a sign of things to come
how AI startups, especially GPT wrappers, can adjust.
Sign of things to come?
If 2023 was about using GenAI as the “tip of the spear” for enterprise AI sales, 2024 is about using GenAI to attack consumer and prosumer app markets. Internationally, the Chinese tech leaders (Baidu, Tencent, etc) had always offered their flagship LLM chatbots for free to begin with. This is the new normal, and AI startups need to rethink their product roadmaps.
It remains to be seen how this impacts the $20/mo “ChatGPT wrapper” industry, which includes Perplexity.ai, etc. I suspect that these apps will grow for at least a few more years, but their pace of growth will slow down significantly as a result of oversupply of free alternatives. Also, they may have to cut prices, or implement a generous free tier just to stay relevant in the “top of the funnel”.
So why is Microsoft and Google giving away chatbots? Well, it’s an old tech industry playbook to give away stuff for free (see Google maps, gmail, AWS, etc), but also there’s some other reasons:
the non-obvious motivation is that inference costs are plummeting, so Microsoft and Google need to find new sources of LLM demand. Improvements in quantization, decoding, and batching techniques, the cost to serve LLMs has plummeted, which made giving away chatbots feasible. So it’s better to shove GenAI onto every consumer touchpoint than to have GPUs sitting around idly. This also implies (ominously) that enterprise demand for GenAI wasn’t strong enough to take up all the GPU supply.
they also get valuable training data by giving away chatbots, which get better with more usage.
It’s also a defensive move, since no one’s charging for chatbots anymore (which is probably why Notion AI decided to make AI features free, also, since they realized that it’s not worth charging for).
Another possibility is that Big Tech is already moved onto promoting “multi-modal” experiences, and text-only chatbots are already old-news. Paying customers are already desensitized to simple Chat UIs, so it now takes a lot more to keep those users from going to other platforms like Perplexity, etc. The multi-modal features will also justify new device upgrades and upsells, or at least that’s the hope. Therefore, it’s just right timing to commoditize the old tech - text based chatbots.
Given that Multi-Modal is a multi-year wave to come, and that Microsoft and Google need to figure out how to utilize their overcapacity of engineers and PMs, I’m willing to bet that Khanmigo won’t be the first app to be offered for free in an industry vertical. Already, AWS has an entire healthcare specific suite of managed services.
Cloud providers will increasingly compete with their ISVs as well as vertical / industry partners.
Implications
Well, I have both good and bad news for startups.
Let me first play the devil’s advocate by saying that this trend won’t kill AI startups or chatgpt wrappers - at least not in the next 6 months. Once a user finds some tool valuable, it often takes quite a lot to convince him/her to try something new, even if it’s free.
People also really underestimate the power of better UX when it comes to acquiring and retaining customers - even if the “core tech” is just a LLM wrapper. Perplexity is arguably a ChatGPT wrapper, etc. If you are targeting techies making $250K/year, then $20/mo isn’t a big deal to them.
What matters more is acquiring the customers effectively in the first place by riding the wave at the right time, which is harder than competing with a free app made by Microsoft with mediocre UX.
That being said, there’s no question that customers are already getting jaded with GenAI, especially when Big Tech is shoving chatbots down your throats. ChatGPT adoption is already slowing, etc. So having a ton of free chatbots available is certainly going to make enterprise AI sales tougher, since the customer will be thinking “I can get GPT-4 for free, do I really need to buy this?”
So what are the immediate implications for AI startups? First, accept that freemium funnel might be non-negotiable for AI startups. You can’t just slap on AI and charge for it, like Notion tried to do. It resets the “baseline” of competition to some “free” version of ChatGPT, or some domain adapted version of it.
Second, AI startups shouldn’t charge more than $20/mo for a product with PLG motion. Of course, feel free to experiment, but my experience has been that it’s really hard to sell anything for more than $20/mo without significant churn these days. Explore steep term discounts or one time payments. This is often a bandaid solution to high churn, but if high churn is the new normal then there’s no other choice.
Lastly, set lower expectations for big exits or financial outcomes. Get used to shorter boom and bust cycles and having limited upside. Or, pivot. If the core of your functionality overlaps significantly with the free GPT-4 powered ChatGPT, then it may not be worthwhile investing into your roadmap.