The US job market seems to be in slow, but steady decline since 2022.
I used the number of job postings on Indeed as reported by St. Louis Fed as a proxy measurement. What’s interesting though, is the number of job postings for Software Development Jobs. I tried to make sense of job market data in a previous issue of this newsletter back in April.
As of now, there are still more or less the same models that could explain the “tech hiring bubble”:
- Covid 2020-2022: Lockdowns and the rise of remote work, vastly accelerated by government stimuli, led to a surge in demand for software developers. This increased wages and number of jobs. Low interest rates led to a boom in Venture Capital, primarily in digital transformation markets.
- Rising inflation (from 2021) and supply chain disruptions (aftermath of primarily the Covid era and the Russia-Ukraine war) led to monetary policy tightening, leading to a decrease in investments of VCs, and thus, to a market correction after over-hiring.
- The mass layoffs starting in 2022 and the decline in tech stocks, along with rising interest rates, led to a cultural shift from growth (“free money”) to profitability (“high interest rates”) in most tech companies.
Times were eventful over the last four or five years, so seeing markets overshoot and correct, is not surprising.
If that would not be enough to stress the markets, there is more to come. According to GitHub CEO Thomas Dohmke, India is projected to be the global hub of developers by 2027.
On top of that, AI plays an important role in developer jobs. With Co-Pilot making developers faster by 55% (according to Dohmke), this would be “an acceleration not seen since the industrial age”. There’s a lot of marketing in this statement. Everything seems to be AI-powered these days. As of now, training AI models and generating responses, requires vast amounts of energy. Markets react to this already heavily. It’s a long time since we saw an “architecture war” in the chip market. The rise of NVIDIA and ARM is a direct result of AI.
In my personal experience, AI coding assistants like Co-Pilot, Cursor, or Pieces, make me more productive. I spend less time writing boilerplate code or handling repetitive tasks. I use Grammarly to check my writing and increasingly use ChatGPT for research. Sometimes, I generate placeholder images with Dall-E or Grok. I do not believe that LLMs are somewhat near “general artificial intelligence” nor that we achieve primate-level intelligence in the foreseeable future.
Given the substantial training costs of AI models — ranging from $4 million for Davinci to $78 million for GPT-4 and up to $191 million for Gemini Ultra — along with the significant energy consumption involved, I see a potential risk for the tech industry. Many of the AI tools we use today, like ChatGPT or Co-Pilot, might become too expensive to offer as widely available services. Larger companies, however, could afford to run their own in-house models. For instance, if the cost to operate a model as powerful as Gemini Ultra is around $200 million, and assuming a software developer earns an average salary of $200k, with a productivity gain of 50%, this would be cost-effective for a team of 2,000 developers. If tools like Co-Pilot become unaffordable or unavailable to individual developers and smaller companies, we could see a growing divide in how technology is utilized between small and large enterprises.
For decades, tech has been an arena where “anyone with a laptop” (along with a brilliant idea and a strong network) could create something impactful. In the future, AI might replace developers—not by eliminating their jobs entirely, but by creating a landscape where the tools that once levelled the playing field are only accessible to the largest players. Jobs may still exist, though on a smaller scale, but individual developers and smaller companies might no longer have access to the same resources as the big players.
Top comments (0)