Roula Khalaf, the Editor of the Financial Times, shares her top picks in this weekly newsletter.
When a CEO starts seeking trillions instead of billions in fundraising, it signals a potential overheating in the sector.
Generative artificial intelligence is poised to revolutionize numerous industries and work dynamics in the long term. However, the news that Sam Altman, the CEO of OpenAI, is exploring an artificial intelligence chip project has sparked significant inquiries.
Sources familiar with the discussions suggest that the project may necessitate raising up to $7 trillion. Achieving even a fraction of this amount, surpassing the combined GDPs of the UK and France, appears quite challenging, to say the least.
Nevertheless, this development underscores the intense interest surrounding AI and its underlying chip technology. The current scenario draws parallels to the dotcom bubble era when soaring valuations and fundraising goals in AI-related stocks evoke memories of the telecom stock frenzy.
During the dotcom era, the anticipation was that the internet would reshape the world, with telecom companies and hardware providers reaping substantial gains. However, the rapid transformation priced into the sector’s valuations proved unsustainable. Similarly, today’s exuberance in AI investments mirrors that optimism.
In the initial internet boom, networking hardware held sway as servers required robust connectivity through routers. Companies made significant investments under the assumption of perpetual server demand growth. This led to a meteoric rise in telecom equipment stocks like Cisco, which surged over 30-fold before the 2000 crash.
The telecom industry’s downfall arrived sooner than expected, transitioning from boom to bust in just four years—outpacing the internet’s societal impact. The oversupply crisis drove over 20 telecom firms into bankruptcy by 2002, causing a sharp decline in stock prices.
In the current AI landscape, chips reign supreme, driving a race among AI firms to secure a larger share of the chip manufacturing ecosystem. As AI models expand in complexity, the demand for chips escalates, intensifying the existing supply shortage.
The sustainability of these chip shortages remains uncertain. Just two years ago, a severe shortage of automotive chips nearly paralyzed the global car industry, only to witness a swift resolution within a year. Today, the auto chip supply has not only stabilized but faces an oversupply in various categories.
The primary risk of rapid cash influx into AI chips is the looming threat of overcapacity. This challenge is already evident in outdated chip segments. With the prolonged sector downturn, Samsung had to curtail production to address a deepening chip surplus, while Kioxia reported a staggering $1.7 billion loss in the last three quarters of 2021. Moreover, over 70 new fabrication plants are in development.
Despite a 14.3% decline in global silicon wafer shipments last year, attributed partly to cyclical chip sector fluctuations and reduced consumer electronics demand, the slump in global chipmaking equipment billings hints at a more subdued growth trajectory than the AI hype suggests.
Another concern is the rapid commoditization of chips. Older-generation chips, like the 40nm variants used in household appliances, were once cutting-edge resources but are now abundant. As equipment depreciates, the prices of outdated chips plummet.
The continuous evolution of chip speed and software efficiency, exemplified by the swift transition from 7nm to 5nm technology in recent Nvidia chips, implies that future chip expenditures may be significantly lower than current projections.
While distinctions exist between the dotcom era and the current AI surge, such as OpenAI’s impressive $2 billion annual revenue and diverse profit avenues, widespread enterprise adoption of AI is still on the horizon. The transformative impact of AI might unfold over a longer timeframe than market valuations and funding forecasts anticipate. Guarding against hype and excessive investments is crucial to avoid the pitfalls of the overhyped 1990s counterparts, underscoring the importance of learning from history.
