Generative AI Revolution: History, Tech Stack, Market Map, Trends

Generative AI Revolution: History, Tech Stack, Market Map, Trends

History of the Generative AI


The global economy used to be driven by industrialization but now it is powered by knowledge and information. This shift has been accelerated by major technological advances such as the internet in the 1990s, cloud computing in the 2000s, and smartphones in the 2010s, which have broadened access to knowledge and transformed the way people communicate, create, and consume content.

AI is predicted to be the next platform shift that will define the “knowledge and information economy” of the future.
No alt text provided for this image
Bessemer Venture Partners, Is AI generation the next platform shift?
No alt text provided for this image

Humans excel at analyzing things, but machines are even better. They can analyze data sets and identify patterns for various purposes, such as detecting fraud or spam, forecasting delivery time, or predicting the next Instagram reels to show. This traditional AI, also known as analytical AI, is becoming smarter every day. While humans are skilled at analyzing, they are also exceptional at creating, writing poetry, designing products, developing games, and writing code. Machines have traditionally been relegated to analytical and rote cognitive tasks, with no chance of competing with humans in creative pursuits until recently. With the emergence of generative AI, however, machines are now becoming good at creating sensical and beautiful things.

??? This new category is called “Generative AI,” meaning the machine is generating something new rather than analyzing something that already exists.

Why is the Time for Generative AI —Why Now?


Sure enough, as the models get bigger and bigger, they begin to deliver human-level, and then superhuman results.

?? Generative AI has the same “why now” as AI more broadly: better models, more data, more compute.

There are several reasons why now is the time for Gen-AI. First, advancements in machine learning and natural language processing have allowed for AI systems to produce high-quality, human-like content. Second, the demand for personalized and unique content in fields like art, marketing, and entertainment has led to an increased need for Generative AI platforms. Third, the availability of large amounts of data and powerful computational resources has enabled the training and deployment of these models at scale.

Generative AI has undergone several waves of development.

  • The first wave (pre-2015) involved small models that were primarily used for analytical tasks but were not expressive enough for general-purpose generative tasks.
  • The second wave (2015-today) focused on scaling up models to surpass human performance benchmarks. However, these models were not widespread. They are large and difficult to run (requiring GPU orchestration), not broadly accessible (unavailable or closed beta only), and expensive to use as a cloud service. Despite these limitations, the earliest Generative AI applications begin to enter the fray.

*A graphics processing unit (GPU) is a computer chip that renders graphics and images by performing rapid mathematical calculations. GPUs are used for both professional and personal computing.

No alt text provided for this image
Sequoia Capital, Generative AI: A Creative a New World

  • In the third wave (2022+), new techniques like diffusion models and better algorithms led to cheaper and more accessible models, with open-source models and APIs becoming available to developers.
  • In the current fourth wave (now), the application layer is ripe for an explosion of creativity as the platform layer solidifies and models continue to improve. The expectation is that killer apps will emerge for Generative AI, just as they did for mobile applications a decade ago.

*Diffusion model is a method for creating images from text prompts. It works by adding random noise to a set of training images, then learning how to remove noise to construct the desired image.?OpenAI's text-to-image model?DALL-E 2?is a recent example.

High-level Tech Stack of the Generative AI Market


In order to comprehend the current state of the generative AI market, it is necessary to establish a clear definition of the existing stack.

The stack can be divided into three layers:

  • Applications?that integrate generative AI models into a user-facing product, either running their own model pipelines (“end-to-end apps”) or relying on a third-party API
  • Models?that power AI products, made available either as proprietary APIs or as open-source checkpoints (which, in turn, require a hosting solution)
  • Infrastructure?vendors (i.e. cloud platforms and hardware manufacturers) that run training and inference workloads for generative AI models

No alt text provided for this image
Andreessen Horowitz, Who owns the Generative AI Platform

Generative AI Market Trends and Open Points


1?? The first wave of generative AI apps are starting to reach scale, but struggle with retention and differentiation.

The first wave of generative AI apps are growing rapidly, with some already earning over $100 million in revenue in areas like image generation, copywriting, and code writing. However, the challenge is not just growth, but profitability and retention. Many of these apps are struggling to differentiate themselves and maintain high retention rates, with some facing low gross margins and difficulty scaling their customer acquisition strategies.

There’s a wide range of gross margins — as high as 90% in a few cases but more often as low as 50-60%, driven largely by the cost of model inference. Top-of-funnel growth has been amazing, but it’s unclear if current customer acquisition strategies will be scalable — we’re already seeing paid acquisition efficacy and retention start to tail off. Many apps are also relatively undifferentiated since they rely on similar underlying AI models and haven’t discovered obvious network effects, or data/workflows, that is hard for competitors to duplicate.

It's not yet clear whether end-user apps are the best path to building a sustainable generative AI business, and there are still many unanswered questions about vertical integration, building features vs. apps, and managing through the hype cycle.

2?? Model providers invented generative AI, but haven’t reached a large commercial scale.

Generative AI owes its existence to Google, OpenAI, and Stability's groundbreaking research and engineering work. Thanks to novel model architectures and heroic efforts to scale training pipelines, we can all benefit from the incredible capabilities of current large language models (LLMs) and image-generation models. However, the revenue associated with these companies is small compared to their usage and buzz. OpenAI dominates with GPT-3/4 and ChatGPT, but relatively few killer apps have been built on their models, and prices have already dropped once.

Countervailing forces, such as open-source models, could compete with closed-source models. For example, we’re starting to see LLMs built by companies like Anthropic, Cohere, and Character.ai come closer to OpenAI levels of performance, training on similar datasets (i.e. the internet) and with similar model architectures. Hosting and commercialization are likely tied, with demand for proprietary APIs and hosting services for open-source models growing rapidly. Model providers face challenges related to commoditization, graduation risk, and whether they should prioritize capturing value.

*A large language model (LLM) is a type of?machine learning?model that can perform a variety of natural language processing (NLP) tasks, including generating and classifying text, answering questions in a conversational manner and translating text from one language to another.

3?? Infrastructure vendors touch everything and reap the rewards.

The generative AI market heavily relies on cloud-hosted GPUs and TPUs, which are the lifeblood of this technology. Infrastructure companies, especially the Big 3 cloud providers (Amazon Web Services, Google Cloud Platform, and Microsoft Azure), are benefiting from the high demand for these resources. ?? It is estimated that 10-20% of total revenue in generative AI goes to cloud providers. Startups training their own models have raised billions of dollars in venture capital, which is also mostly spent with cloud providers. Nvidia is the biggest winner in generative AI, having built strong moats around its business via decades of investment in GPU architecture, a robust software ecosystem, and deep usage in the academic community. The infrastructure layer is a lucrative, durable, and seemingly defensible layer in the stack, but cloud providers need to address key challenges such as holding onto stateless workloads, surviving the end of chip scarcity, and the possibility of a challenger cloud breaking through.

*Tensor Processing Units (TPUs) are?Google's custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads. TPUs are designed from the ground up with the benefit of Google's deep experience and leadership in machine learning.

Generative AI Models and Applications?


No alt text provided for this image
Sequoia Capital, Generative AI: A Creative a New World

The diagram above illustrates the model layer that will serve as the foundation for each of the categories. This layer comprises various models with varying degrees of complexity and training, depending on the specific application area. These models are designed to generate outputs that can be further refined and used in a range of applications.

For example, in the case of text, we can expect models that are proficient in generating short to medium-form writing, which could be utilized in various applications, such as content creation, chatbots, or even automated translations. Similarly, in the case of code generation, models can be trained to generate code snippets, complete functions, or even entire programs, which could significantly improve developers' productivity.

Overall, these model layers provide the building blocks for various applications in their respective categories, and as the models continue to improve, we can expect to see new and innovative uses of these technologies.

No alt text provided for this image
Sequoia Capital, Generative AI: A Creative a New World

Market Map of the Generative AI


The market is bursting with innovation as we see new startups emerge every day! In fact, the recently announced,

YC W23 Batch of 254 startups includes a whopping 54 startups leveraging "Gen-AI" technologies! (See here)

?? It's clear that the future is bright for generative AI, and I am thrilled to be a part of this cutting-edge field.

Check out the latest generative AI market map, created by Sequoia Capital and our very own Gen AI investment Coqui ?? .

Let's keep pushing the boundaries and creating a world powered by AI!

No alt text provided for this image
Sequoia Capital

Also, here you can find Antler’s Gen-AI Landscape list with the details of funding.?

How Will the Generative AI Market Evolve Over Time?


?? It is uncertain whether there will be a dominant player in the generative AI market in the long run, as the current data does not provide a clear indication.

  1. Applications lack strong product differentiation because they use similar models;
  2. Models face unclear long-term differentiation because they are trained on similar datasets with similar architectures;
  3. Cloud providers lack deep technical differentiation because they run the same GPUs; and even the hardware companies manufacture their chips at the same fabs.

The Future of Gen AI: Who Will Lead the Way?


The most successful teams in the Generative AI market, will prioritize exceptional user engagement and use it to enhance model performance through prompt improvements, model fine-tuning, and utilizing user choices as labeled training data.

By leveraging their great model performance, they can attract and retain more users and continue to drive engagement. Instead of trying to be everything to everyone, these companies will likely focus on specific problem areas such as code, design, or gaming. They will begin by deeply integrating into existing applications to gain leverage and distribution before eventually attempting to replace incumbent applications with AI-native workflows. Although it will take time to develop these applications and accumulate users and data, we believe that the best companies will be able to endure and potentially become industry giants.

No alt text provided for this image

PS. This piece was co-written with GPT. We hope you enjoy reading it as much as we enjoyed creating it! ???

Resources


Below, I have listed all the resources that inspired and assisted me in writing this article.

G?k?e Duman

Co-Founder & Managing Partner @ Ingosa | Revolutionizing digital marketing with conversational ads and generative AI-driven solutions

1 年

Thank you Ba?ak, for sharing such an insightful and informative article on the latest developments in generative AI. The market map you provided was particularly helpful in identifying key players in the space. I look forward to reading more of your work in the future!

Erdal Bektas

Gen-AI for inventors to get patents directly

1 年

We are in the gen-ai space and always would love to chat.

回复

Great article Ba?ak! Always interesting to read about Generative AI.

Ahmet Bilgen

"Global Tech Visionary | Co-founder of: Trendbox | Successful Exit Foriba(FIT Solutions) | Endeavor Entrepreneur | Angel Investor | Empowering Innovation | Building the Future Together"

1 年

I never said Cripto is going to change the world!

Haluk Aykul y.

Entrepreneur, Investment Advisor, (M&A Advisory & Startups) Expert on Executive Search, Advisor for creating New Businesses, Career Advisor

1 年

Excellent sharing Basak , thanks for sharing. Hard task to predict the future of landscape yet a big opportunity for those who would make the right moves. An interesting trend would be to be integrators for big players that are extremely generous to allocate ressources in this new era

要查看或添加评论,请登录

社区洞察

其他会员也浏览了