AI is thirsty.
Tech giants are betting on nuclear power as the AI wave submerges mainstream business – is it the only viable alternative?
Many would say yes, but time will tell when the dust settles on the AI megahype. What is clear right now, is the extreme cost of training today’s AI models.
The growing complexity of today’s models is pushing businesses to search for solutions further afield. Trust in SODA has collaborated with our partners at DeepRec.ai to bring you a deep dive into the atomic dreams of big tech.
The Economics of Training Data
Estimates suggest AI is already consuming more power than a small country. The IEA expects global electricity demand to rise at an average of 3.6% through 2026, with 85% of the demand coming from outside advanced economies (crypto’s anticipated move to mainstream finance will likely exacerbate this demand).
Training data sits at the heart of AI development. It’s linked to rising costs through a complex web of interrelated factors, including:
- Europe’s high energy costs impede competitiveness, forcing pioneers to reconsider where they build their data centres. Rising demand for GenAI is tipped to make the situation worse.
- Complex data sets are needed to develop increasingly advanced AI platforms. They’re costly to align with specialised use cases like autonomous cars or medical imaging, and they’re even more costly to maintain in the long term.
- Regulatory compliance is piling on the pressure. Between the EU AI Act and Trump 2.0, the regulatory space is undergoing some dramatic changes, and data security is at the epicentre. Plus, it’s regulatory red tape that’s bottlenecking progress on nuclear power projects.
Energy links it all together. Resource-intensive processes are well-suited to the stable, 24/7 output provided by nuclear power centres. Google, Amazon, and Microsoft are planning to invest heavily in the coming years, arguing that the move presents the low-carbon alternative needed to meet growing demand.
There’s not much info (yet) on just how much energy GenAI uses, and some critics of the nuclear approach suggest that big tech is jumping the gun.
Others are claiming it’s the only sustainable, economically viable route in an era of unprecedented energy consumption, provided organisations can navigate a complex regulatory environment.
As the Financial Times points out, the prospect of environmental deregulation in the US creates space for a ‘rare bipartisan alignment.’ Could nuclear energy bridge the divide? Would deregulation lead to greater opportunities for sustainable energy solutions in the future?
AI has made some big promises in this space, but it needs to be given the tools to get there. With the cost of developing models growing at a factor of around 3x in the last year, the largest models now cost upwards of a billion dollars.
If an alternative isn’t found, the growing focus on high-quality training data sets will price SMEs (small and medium-sized enterprises) out of the market, leaving only those with the biggest budgets the chance to compete.
On the other hand, rising costs have spurred progress in architecture optimisation and synthetic data, creating an opportunity for pioneers to compete without needing the same scale as big tech.
What Does it Mean for Data Jobs?
We can expect demand for most data-focused roles to grow alongside the adoption rate of more sophisticated AI models. We’re likely to see the talent market evolve in some peculiar ways over the next few years:
- An uptick in energy-adjacent data roles – the demand for professionals who can help close the gap between technology and energy infrastructure looks set to rise.
- The impact of reshoring and nearshoring – between geopolitical tumult and regulatory pressures, an increase in reshoring and nearshoring could disrupt data-specific roles as companies look to centralise their workforces.
- Emerging markets – Niche domain expertise is on the rise as adoption picks up.
- Increased automation – stronger AI systems will result in more use cases for automation. Some leaders (including those from our latest networking event in Zurich) suggest that this only creates more space for employees to use their brains.
- The convergence of compliance and data – A global uptake in AI-specific legalisation will drive demand for data governance roles.
- Architecture – as energy efficiency rises to the top of the priority list, organisations of any size will need experts to design efficient architecture if they plan on developing AI models.
Exciting Times Ahead
Innovations in nuclear energy represent the chance to build a more sustainable future for the energy-guzzling AI systems we’ve gotten friendly with over the last few years, and we expect it to drastically reshape the labour market.
How are you planning to approach the rising costs of AI development? Our talent consultants would love to hear from you.
Contact us directly to join the conversation:
Francis Alexander – Data specialist at Trust In SODA, Germany
James Davis – Data science and machine learning specialist at DeepRec.ai, USA