The writer is founder of Sifted, an FT-backed site about European start-ups
The latest Californian gold rush is about a mad scramble over generative AI. The big US tech companies, such as Google, Microsoft, Meta and Palantir, as well as a swarm of venture capital firms, are all manically digging for new seams of digital treasure. But the big, and as yet unanswerable, questions are: who’s going to end up spitting dust and who will bag the most gold?
The obvious guess is that the big companies that are developing these text, image, video and audio generation models will dominate the field. As Lina Khan, the chair of the Federal Trade Commission, has written, a handful of powerful businesses appear to control all the necessary raw materials: vast stores of data, computing power and cloud services. To which, she might have added: they also have many of the world’s leading AI researchers and mountains of cash.
“The expanding adoption of AI risks further locking in the market dominance of large incumbent technology firms,” Khan recently wrote in The New York Times. Such a worldview provides more traction to Khan’s trustbusting drive.
But that is not how the world looks to those inside some big tech companies, judging by a leaked memo from one Google executive entitled We Have No Moat. The executive noted this April that Google and OpenAI, heavily backed by Microsoft, may have developed the most capable, closed generative AI models, such as Bard and GPT-4. But they were already in danger of being outrun by more agile competitors that were building smaller, cheaper and more customisable open-source AI models and luring away some of Google’s best researchers. “The uncomfortable truth is, we aren’t positioned to win this arms race and neither is OpenAI,” the executive wrote. “Who would pay for a Google product with usage restrictions if there is a free, high quality alternative without them?”
This week, Google tried to up its game by announcing it would embed generative AI in a broadening range of services, playing catch-up with Microsoft. But, according to the Google memo, the main beneficiary of the trend towards open-source models might be Meta, which has also been pivoting towards AI. Having launched its own open-source LLaMA model in February, Meta is now aiming to build the platform on which others can play. Just as Google constructed a new app ecosystem around its open-source Android mobile phone software, so Meta could emerge as the platform on which innovation happens. “The one clear winner in all of this is Meta,” the Google executive wrote.
Meanwhile, VC investors are betting that a fresh wave of generative AI start-ups, including Anthropic, Cohere, Stability AI, Inflection and AI21 Labs, can also successfully pan for gold. Their logic is that at least some of these start-ups can move faster than the bigger companies, dominate select market niches and largely ignore costly safety controls (something which should alarm the regulators).
The future will belong to smaller, specialist generative AI models that are cheaper to train, faster to run and serve a specific use case, says Yoav Shoham, co-founder of the Israeli start-up AI21 Labs. “The moat is not technology. It is the relationship with the consumer,” Shoham tells me. “I think it will be a ‘few-takes-the-most’ market, not a ‘winner-takes-all’ market.” Scores of other start-ups, which only provide a generic service and have no traction with customers, will fail.
Established incumbent companies in most industries that can feed their own proprietary data into generative AI models and fine tune the outputs are also likely to thrive. The challenge for them is to re-engineer their organisational structures to exploit the new technology. The other certain winners will be the “picks and shovels” companies that provide the tools for this technological transformation. The big cloud computing companies, AWS, Google and Microsoft, will profit from the models’ voracious demand for computing power. But Nvidia, the dominant designer of graphics processing units that power most AI models, also stands out. “We are at the iPhone moment for AI,” says Jensen Huang, Nvidia’s chief executive.
Such is the speed at which the sector is evolving, however, that today’s best guesses may turn into tomorrow’s shredded betting slips. The history of other general purpose technologies, such as electricity, motor cars and the internet, suggests that generative AI will create new markets and business models that no one can imagine today. It may well be that a company yet to be founded mines the most gold.