This article is part of our series that explores the business of artificial intelligence
Making predictions is not my forte. But it seems like I got it right last year when I said that Anthropic would soon gravitate toward one of the big tech companies. This week, Financial Times reported that Google is investing $300 million in the San Francisco–based artificial intelligence lab.
The details of the deal between Google and Anthropic are scarce. But it bears many similarities with the partnership between Microsoft’s investment in OpenAI, which recently got a multi-year, multibillion-dollar extension.
With an escalating arms race for the domination of the next major shift in computation, AI research labs like Anthropic will be hard-pressed to avoid being sucked into the power struggle between tech giants. And hanging in the balance will be the delicate equilibrium between pursuing scientific endeavors and creating monetizable technology.
What is Anthropic AI?
Anthropic was founded in 2021 by Dario Amodei, the former VP of research at OpenAI. The company does not publicize its work aside from publishing research papers. But it has been very successful in raising funding for its research. The company raised $124 million in series A upon its founding. Last year, it added another $580 million in series B. And now it has another $300 million from Google to continue its research.
Anthropic’s website describes it as “an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems.” The company seeks to make progress on the problem of today’s “general systems” being “unpredictable, unreliable, and opaque.”
The website lists more than a dozen research papers on topics such as large language models (LLM), reinforcement learning from human feedback (RLHF), large generative models, transformer models, and interpretability. Their last paper was published in December 2022.
The company has no public plan for launching commercial products and is focused on research. But they state that “down the road, we foresee many opportunities for our work to create value commercially and for public benefit.”
Google’s interest in Anthropic
Anthropic has selected Google as its preferred cloud provider to “train, scale, and deploy its AI systems.” According to Amodei, Google Cloud will support Anthropic’s next phase, which is “to deploy our AI systems to a larger set of people.”
The company is developing an AI assistant called Claude, which has not been released to the public (yet some media outlets are already comparing it to OpenAI’s ChatGPT). Claude leverages “Constitutional AI,” a technique Anthropic scientists designed to supervise AI systems with other AI. The technique aims to “make it possible to control AI behavior more precisely and with far fewer human labels.”
Why would Google be interested in partnering with Anthropic while it already has some of the most talented AI scientists at Google Brain and DeepMind? I think there are three key advantages, all of which run in parallel with Microsoft and OpenAI’s partnership.
Well, for one thing, Microsoft’s investment in OpenAI has shown the benefits of investing in an AI lab without acquiring it.
Despite the excitement surrounding it, generative AI is an emerging field. We are still exploring its ethical, legal, and technical challenges. Lawyers are still trying to figure out how models such as DALL-E and Codex affect copyright infringement laws. The developers of GPT-3 and ChatGPT are constantly creating new safeguards to prevent the LLMs from generating hate speech and other harmful content. Schools and universities are concerned about students using generative models to cheat in exams.
For large tech companies such as Google and Microsoft, any of these concerns can turn into a PR nightmare and have a direct impact on their brand and reputation. Smaller companies such as OpenAI and Anthropic have more maneuvering room to make mistakes and learn.
The second benefit of the partnership is Anthropic becoming a long-term customer of the Google Cloud Platform. Google’s investment was done in GCP credits, not in cash. This will potentially turn Anthropic into a long-term high-paying client for Google.
Finally, working with Anthropic will enable Google to learn to improve its AI hardware to run models at scale. Google already has a seasoned AI hardware engineering team, and its TPU processors are among the best-in-class inference engines. But the Anthropic team was involved in OpenAI’s “AI & Compute” research efforts, and it can surely bring new ideas to the table.
The future of Anthropic’s AI research
The question that remains to be answered is where the partnership leaves Google and Anthropic. I think Google is the bigger winner. At the very least, it will get Google a high-paying customer and prevent a promising research lab from being absorbed into Microsoft’s growing share of the market for applied artificial intelligence. In a best-case scenario, Google will be looking at a future acquisition and more AI firepower to compete with other tech giants.
What about Anthropic? I’m afraid they will get the short end of the stick. For the moment, they get a lifeline to continue their research. But they will find themselves locked into Google’s cloud platform because they are receiving their investment in GCP credits. Once their scalable AI infrastructure is in place, it will be very difficult and expensive to switch to a different provider. And given their commitment to expanding their operations, I wouldn’t be surprised if their $300 million cash infusion dries up within the next year or so. After that, they will have to raise more funding to pony up for their Google Cloud costs. If they’re lucky, Google will grant them more GCP credits. Alternatively, they will have to look for channels to monetize their technology. And who better to do it for them than Google, their cloud provider?
There are too many similarities with the trajectory that OpenAI has taken since it began getting investments in Azure credits from Microsoft. While OpenAI continues to publish amazing research, it has also become beholden to the commercial interests of its main financial backer. It is effectively focusing more on research that aligns with Microsoft’s line of products and can be monetized faster (language models, text-to-image generators, source code generators) and less on areas that have a pure scientific purpose with no short-term monetization in sight (robotic hands, game-playing bots, evolutionary algorithms, etc.).
Before founding Anthropic, Amodei and other researchers reportedly left OpenAI over disagreements with the company’s direction after taking a $1 billion investment from Microsoft. They were concerned that with Microsoft’s investment, OpenAI would steer toward a commercial path and detract from its original focus on AI safety.
Now, Anthropic has taken the first step toward a similar path. Will the research lab manage to stay true to its scientific goals while raising money from big tech? Or will it end up becoming another de facto arm of Google, creating tech to be integrated into Google’s products? And what will happen to the other AI labs that are hoping to carve a different path toward artificial general intelligence?