2024 was another monster year for AI progress and spending on AI. In 2024, organizations worldwide spent an estimated $235 billion on AI, encompassing applications, infrastructure, and related services.
The Future of AI is not certain but the huge spending is for the belief that we will transform the information industries with AGI-like or even super intelligent capabilities. There is evidence that AI capabilities are providing PHD level answers and reasoning from the OpenAI O3 model and from Google’s Deep Research system. Elon Musk told Mark Penn at CES that in 3-4 years we could perform all human labor tasks and can do any cognitive task.
There is visibility where we know there is growing usefulness for AI. There is value in answers that replace a hundred billion dollars worth of Google search. There is value in creating images and video. There is value in improving a trillion dollars in human programmer productivity. Revamping the $5 trillion global IT industry is a clearly achievable objective. There will be new products and uses.
Brad Gerstner, Altimeter Capital, says that the industry is still compute and energy constrained in 2024 and will remain compute and energy constrained at the end of 2025. There is also a memory shortage.
The core of the AI hardware story of 2025 is 2+ million Nvidia B200 and B300 chips. The ASIC chips will matter but will be tiny in the overall market.
The improved reasoning inference need perhaps 100 times more energy to get the quality answers. OpenAI O3 using $3000 per answer is about 3000 GPU hours for an answer.
Agentic AI requires agents to perform dozens or hundreds of steps and actions.
More reasoning steps, more complex reasoning and more agentic steps and actions all require more compute and energy.
The clear evidence is that AI performance and infrastructure is still scaling in a sustainable way. AI pre-training may be having a slowing curve but post-training and test-time training are just starting.
Bill Gurley said all incremental free cash flow is now going to building and developing AI. The forecasted cumulative incremental free cash flow of major companies would be an approximation of global AI spend. AI investments will need to return increased free cash flow and corporate returns to enable continued spending and revenue growth.
In 2024, generative AI accounted for 17% of global AI spending, and this share is projected to rise to 32% by 2028, with about 60% five-year CAGR. Bloomberg Intelligence estimates that generative AI spending will grow by 71% in 2025.
Real world AI is emerging with self-driving cars and trucks which could see inflection moments in 2025. There could be safer than human driving cars as soon as Q2 of 2025. Humanoid bots will start scaling at a few tens of thousands in 2025 and over a hundred thousand in 2026 and perhaps a million in 2027.
Amazon is spending $11 billion plan to expand its cloud computing and AI infrastructure in Georgia. Microsoft announced a plan to spend $80 billion on AI in 2025.
Sequoia Capital reports the AI industry spent $50 billion on Nvidia chips for training in 2024 but generated only $3 billion in revenue in 2024.
Nvidia’s AI spending is primarily focused on R&D and strategic investments. In 2024, Nvidia invested $1 billion in various AI ventures, including funding rounds for AI startups and corporate deals. Nvidia’s R&D expenses have been steadily increasing, reaching $8.7 billion in fiscal year 2024.
Nvidia has applied AI to improve its own chip designs.
There are larger projects to use AI to speed up and improve science and technological development.
In 2023, TSMC spent over $13 billion on R&D and planned to increase its R&D budget for 2024 by as much as 20%. TSMC’s capex spending in 2024 was projected to be between $28 billion and $32 billion and this will increase to $36-40 billion in 2025.
Here are estimates of the major models that have and will be released and the level of compute that will be used. This is based upon information and statements made by executives and other sources.
There is other analysis of what will be needed in the changing AI emphasis on test time training and inference. There will be new chips and systems focused on different aspects of inference.
There is an estimation of the scaling constraints. However, there will be thousands of brilliant researchers and companies that will remove or mitigate constraints.
AI is worldchanging and the level and rate of change is more likely to be larger than expected instead of less than expected.
The following newsletters will dig into the details of approximating usage of infrastructure for different companies seeking utility from their AI spend.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Photonics, quantum computers or some other neuromorphic substrate/framework will be the key to really supercharging this tech by massively dropping the energy consumption in the training stage. That’ll be what finally brings AI in line with biological systems and their level of efficiency. Jensen Huang will be pivoting that way at some point behind-the-scenes, if not already. Most people have no idea what is coming because they fail to take this into account, for a start. In the meantime, improved GPUs and training will still be revolutionary but more like dial-up versus broadband and fibre.