Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Is the AI bubble deflating, or only just getting going? Nvidia shed nearly half a trillion dollars in stock market value in the four trading sessions after its latest quarterly earnings, even though it comfortably beat official expectations.
Add in Apple, Microsoft, Alphabet, Amazon and Meta, which have dominated the stock market indices since AI mania took off, and the pullback over the past two months has cost investors $1.8tn. Amid mixed economic signals and signs of a stock market rotation into other sectors, AI hype has no longer been enough to carry the market before it.
But hope springs eternal. In the last few days, it has emerged that OpenAI is looking for a valuation of more than $100bn in its latest fundraising — and potentially much more. Despite a crowded field of companies building AI models, new start-ups are still attracting breathtaking valuations.
And despite the summer setback, most of the gains from the AI boom are still intact. The six biggest tech companies have still seen their combined market cap grow by nearly $2.9tn, or 27 per cent, since the start of the year.
Still, a mood shift has occurred, and volatility looks like becoming the new reality, Nvidia is still delivering stellar results by any normal yardstick, but the period of shock and awe that transfixed Wall Street over the past year is past and a degree of sobriety is returning.
A pause was also likely at this point as the AI race ran into business reality. After the huge burst of capital spending that began early last year, carrying infrastructure companies like Nvidia higher, finding economically valuable uses for all the new AI capacity was always going to take time.
The narrowness of capital spending has also made the gains feel more precarious. Just three big companies accounted for 49 per cent of Nvidia’s accounts receivables at the end of its latest quarter.
Yet the eye-catching action in the private market suggests that this is still a technology in its infancy. Even as some recently hyped AI start-ups have foundered, new ones are rapidly emerging to take their place. This has nowhere more apparent than in the market for large language models and other “foundation” models on which the generative AI boom has been built.
The risk for companies in this market has been that foundation models are becoming an undifferentiated commodity and that inherent flaws in the technology, like its propensity to hallucinate, will severely limit its uses. If so, then whatever pricing power companies have had will evaporate and a brutal consolidation will set in.
Some have already shut their doors. The leaders of Character. AI, which a year ago was reported to be trying to raise money at a $5bn valuation, have been absorbed into Amazon. Inflection, once valued as much as $4bn, has been absorbed by Microsoft.
But the latest funding news shows that there is still plenty of money in Silicon Valley backing an alternative view. Partly with Nvidia’s financial support, the Japanese start-up Sakana AI was valued at more than $1bn in its A round this week. Safe Superintelligence, partly led by OpenAI co-founder Ilya Sutskever and with only 10 employees, has just been valued at $5bn by investors including Sequoia and Andreessen Horowitz.
Sakana is a reminder that the proliferation of models has barely begun. Around the world, pressure to train local models using local data, while keeping them under local control, is intense. Also, while big models have delivered the biggest technical advances, they are being superseded in many practical situations by a much larger number of smaller models that have been honed with specific data relevant to the task at hand.
Meanwhile, Safe Superintelligence, as the second part of the company’s name suggests, is a bet that a much bigger AI prize lies within reach: a level of artificial intelligence that far surpasses both the current state of the art, as well as topping human intellect.
The start-up’s founders claim they can see a better way to reach what they call “the top of the mountain” in AI. What that will entail is not clear — though it comes amid a fervid search among AI researchers for new and more efficient ways of training models that threaten to soak up exponentially more data and energy, as well as new techniques that will overcome the limitations of the current technology.
Wall Street is facing a period of AI exhaustion. But judging by the top venture capital names lining up behind Safe Superintelligence, there is plenty of room for much more AI hype to come.
