AI Special: If it’s free, the planet may be the price
If AI is going to transform climate technologies and accelerate decarbonisation it'll first have to prove profitability in consumer applications
We’re in the eye of the AI storm right now and, like previous tech hype cycles, it’s easy to get carried along on the sweet smell of potential. But maybe we can we learn to spot signals from previous cycles? Amidst talk of corporate climate goalposts shifting to accommodate AI’s energy intensity, there’s less recognition of another warning sign flashing here. That is the remarkably slender cost of AI to the consumer. If it’s (practically) free, what exactly is the price?
A Carbon Catch-22
Microsoft's about-face on its ambitious carbon goals - to remove more than it emits by the end of the decade - stalled as AI demand increased. Emissions ramped up 30% higher than in 2020. President Brad Smith and founder Bill Gates are now telling a new story of short-term pain for long-term gain. They're betting AI's future breakthroughs in energy efficiency and clean tech will outpace today's emissions surge.
This depends on a realisation of the breakthroughs the technology promises.
The AI market faced a reality check in September. Nvidia, the dominant AI chip supplier, saw $279 billion in value evaporate – the biggest one-day drop in U.S. stock market history. Investors were waking up to the fact that AI's promised revolution hadn't spread beyond Big Tech's echo chamber. As one strategist put it, "We haven't really seen AI spread out across the economy.
None of this improves the odds of Microsoft making up lost time on its carbon goals.
The Hidden Cost of 'Hello, ChatGPT'
While AI training grabs headlines, it is everyday use that's the real energy hog. The 'inference phase' – when you chat with AI – gobbles up 90% of AI's energy costs. Each query burns a quarter of the power needed to boil a kettle.
You may or may not be paying for the latest models. I use Claude 3.5 at the very agreeable monthly sum of £18 and am in the habit of putting it through its paces to help with tasks on and off throughout my working day. I make a lot more queries than I make cups of tea. And I drink a lot of tea.
If you’re wondering if the price we’re paying for AI chatbots is sustainable at its current rate, you’re not alone. A dissertation-length tirade from blogger Edward Zitron titled ‘The Subprime AI Crisis’ was pounced upon by AI naysayers earlier this month and spread panic about the sustainability of big tech valuations based on the nebulous promise of future growth for an industry that, in the short-term, is “burning money”. Both on unrealised big tech investments and nascent products and services - like ChatGPT and Claude - that are losing money every time they’re used:
OpenAI’s models and products… are deeply unprofitable to operate, with the Information reporting that OpenAI is paying Microsoft an estimated $4 billion in 2024 to power ChatGPT and its underlying models, and that's with Microsoft giving it a discounted $1.30-per-GPU-an-hour cost, as opposed to the regular $3.40 to $4 that other customers pay. This means that OpenAI would likely be burning more like $6 billion a year on server costs if it wasn’t so deeply wedded to Microsoft — and that's before you get into costs like staffing ($1.5 billion a year), and, as I've discussed, training costs that are currently $3 billion for the year and will almost certainly increase […]
Those of us whose work has already been transformed and accelerated by AI will rightly take issue with Zitron’s claim that “it isn't clear whether generative AI actually provides much business value at all”. But that doesn’t help us square our use of a product that, right now, looks unsustainable.
Breakthroughs or Pipe Dreams?
Where is the evidence of Microsoft’s promised accelerating sustainability curve?
The World Economic Forum is looking to the technology to offer breakthroughs in local and global climate modelling, to accelerate innovation in prototyping for materials science for better, cheaper batteries and solar cells.
I ask Claude 3.5 who, ever the optimist, echoes these sentiments. But dig deeper, and many of these 'breakthroughs' amount to marginal optimisations. Are we grasping at straws or on the cusp of true innovation?
Sam Altman and Microsoft have big bets on nuclear fusion startup, Helion Energy. AI has a key role in enabling the new generation of nuclear technologies - from managing the vast data needed to ensure stable fusion reactions to predictive maintenance, essential to safety.
But with conventional plants taking a decade to build, can this tech scale fast enough to meet climate goals? The promise of small modular reactors (SMRs) and their much faster turnaround time looms, but regulatory hurdles will be a formidable obstacle.
The Three Levels of AI Adoption: From Cost-Cutting to Revolution
Azeem Azhar's framework for AI adoption provides a useful guide for how AI might meet its loftiest goals.
Here level 1 is ‘do what we do cheaper’, automating routine tasks etc. Level 2 is ‘do what we do, just do it better’, which pretty well describes what’s happening in all the examples of accelerating sustainability we can point to above.
Level 3 is ‘do entirely new things’ but Azhar calls out the challenge of realising this kind of innovation: it will take big shifts in process within the organisation. And we should add to this in the case of energy that it may also require big shifts in regulatory thinking, something even more culturally inflexible than the corporation.
For Microsoft to bend that emissions curve, it needs to master – and teach – these seismic shifts.
Meanwhile, the heat is on for AI to prove it can turn all those Nvidia chips into money. If Apple can finally unlock a role for AI in consumers' lives - when it stops stalling on the rollout of Apple Intelligence - they'll get their return from still more hardware sales (a sustainability topic for another note). That’s just a short-term fix for the economics: there are numerous routes to making consumer AI pay down the line, from the inevitable advertising model to entirely new interfaces for retail.
We're probably in a golden age of consumer and small business AI experience, enjoying a subsidised race for audience scale ahead of all the messy business model experimentation. But realising the big promise of the technology - including bending that emissions curve against tight timelines - hinges on sustainable monetisation.
Without a clear path to profitability, the massive investments needed to achieve AI's lofty goals may never materialise. The race is on not just for technological breakthroughs, but for viable business models that can sustain AI's ambitious journey.
References:
Microsoft’s AI Push Imperils Climate Goal as Carbon Emissions Jump 30%, Bloomberg
Nvidia $279 billion wipeout — the biggest in U.S. history — drags down global chip stocks, CNBC
Why did Nvidia suffer a record $279 billion loss, the biggest for a US company? First Post
The Subprime AI Crisis, Edward Zitron
Altman’s $3.7 Billion Fusion Startup Leaves Scientists Puzzled, Bloomberg
Microsoft Makes Its First Nuclear Fusion Deal, Bloomberg
About 33_Zero
33_Zero works on brand and comms strategy with businesses that recognise unprecedented change needn’t be a threat but can be an opportunity. We help you understand what the changes mean for your customers, for your place in the culture and for the way you create and communicate value. And we plan ways for your brand to show up and participate in this opportunity.
Email jamesp@33seconds.co to find out more.