2023 AI Players and Predictions

Sridhar Ramaswamy and Vivek Raghunathan on 01/13/23
With core expertise in applying AI to search, using existing AI technologies, and training our own models, here is our analysis of the industry landscape and our 2023 predictions for what is to come.

2022 will be known as the year when a lot of AI technology became accessible to hackers and developers. With the recent explosion of capabilities as shown by Chat GPT and Stable Diffusion, it is clear that AI applications to our daily lives are going to explode in 2023. This could be one of the most exciting technological advances since the iPhone.

What’s harder to do is to make sense of the companies best positioned to take advantage of this breaking dam, what the market looks like and where it is likely headed in 2023. With core expertise in applying AI to search, using existing AI technologies, and training our own models, here is our analysis of the industry landscape and our predictions for what is to come.

Let’s start off by categorizing the AI players into a 5 classes.

  1. Foundation model players
    These include Anthropic, OpenAI, Cohere, and AI21. All of these players will all have instruction following models that are more or less substitutes for each other. Their businesses will all be based on an API model. Their APIs will all converge to (a) generations (b) embeddings (c) fine tuning (d) classification. They will all attempt to recruit AI frontend startups (see below). Inspired by the likes of Stripe, they will use an “index YCombinator” business model, and outsource GTM to AI frontend startups. The moat for these companies will be the quality of their model,  their ability to recruit loyal customers who use their APIs.

  2. AI frontend startups
    These are companies like Jasper and Copy.AI that are vertical specific “prompt engineering layers” on top of the foundation model players. To a first cut, all these companies will look like a designer, 2 full stack engineers, 1 BE engineer, 1 applied ML engineer and a GTM team. These companies will use APIs from the foundation model players to stitch together end user solutions. Their strength will be user focus, and a relentless focus on GTM. Most of YCombinator’s 2023 vintage will be companies of this form. A large number of these companies will be marketing tech companies, and the market will be flooded with different styles of textual and visual marketing artifacts because the cost of generation of tailored messages is phenomenally lower than it used to be. These companies will look a lot like a dressed-up version of the enterprise SaaS companies from the last decade (PLG, consumer grade UX), with an “AI intelligence layer”.

  3. Copilot/LLM for X companies
    These will include the original CoPilot for programmers (GitHub/MSFT), “LLM for Chatbots” (Character.ai, Inflection), “LLM for RPA” (Adept), “LLM for search” (Neeva) and many others being incubated as we speak. These will also include OG LLM companies like Cresta (enterprise chatbots) and Lilt (translation). All these companies will make a bet that task-specific LLM tech + deep understanding of user + product is how the market is won. They will need big capital raises to pull off their mission.

  4. Tooling companies. These are classic shovel providers in a gold-rush and will include the labeling companies (Scale.AI, SurgeHQ, Snorkel), training infra companies (MosaicML, Stronger Compute), inference infra companies (Goose.AI), and many others.

  5. Big compute clouds like GCP, AWS, Azure and Oracle
    These companies will all wake up to the business opportunity that LLMs provide and compete very aggressively to go up the value chain beyond being just “GPU compute providers”. Azure threatens to have run away with this game already.

Over the next year, we expect to see multiple battles, consolidation, and splintering across these categories. Here are our top predictions of what 2023 looks like:

  • Compute clouds and foundation model players will collide. The compute clouds will all move towards making “LLM AI capabilities” a feature of their platforms. This will cause the first and last categories of companies to converge. Microsoft already has its OpenAI investment. Google will join the rungs of the foundation model providers by exposing an LLM API via GCP. Amazon and Oracle will buy out one of the independent foundation model companies, or roll out their own models. It’s an open question if there will be any Snowflake and Databricks-style independent companies at the end of 2023.

  • The CoPilot for X”companies will find themselves competing with AI frontend companies that use foundation model APIs to beat them to market. We’ll see this play out in search, RPA, programming and other categories. The jury will be still out on whether you need to invest in deep tech to succeed in these verticals, or can build a frontend business on top of the pure foundation API players. We think it’s the former.

  • The tooling companies will find the going a bit hard. The largest of the foundation models will turn out to be great for producing labeled data, often better than humans. The AI labeling companies will find themselves competing with the large models they helped bootstrap. The infra companies will find the open source tooling continues to get better and the GPU compute availability situation becomes easier, eroding their advantage. Both these classes of companies will react by “moving up the value chain” to solve for business use cases and find themselves in competition with the “AI frontend companies” and the “CoPilot for X” companies.

  • OpenAI will try to be both an API provider and a “CoPilot for X” company for many use cases. This will lead to channel conflict and friction with its partner companies. We will start to see beginnings of rent seeking behavior.

  • And finally, traditional enterprise companies like SalesForce and ServiceNow and UIPath will find themselves at the end of an onslaught the likes of which they haven’t seen before. They will be threatened by one of the “Copilot for X” or “AI frontend” companies who will promise automation to disrupt their existing business model. They will react by acquiring “CoPilot for X” companies.