The AI Endgame: Ads for the masses?

Published Wednesday 21st January 2026

The AI industry could be in bubble territory with enormous capital spending on data centers and infrastructure but lethargic revenue and profit that can't justify the outlay. Read on to understand why OpenAI's announcement of advertising in ChatGPT is likely to be a battle for a grand prize and how marketers can get ready for a new channel.

Maximum Hype

We have all heard of Gartner’s ‘hype cycle’, the framework that describes the maturity, adoption and social perception of new technologies. Simply put, according to the theory all new technologies go through a roller coaster of positive and negative public perception over time.

There is no doubt that whilst its dependent on sector, country and type; public perception of AI has risen over the 2020s¹. Despite the odd blip, such as a decline of perceived value in generative AI last year,¹ there is absolutely no doubt 2025 was peak frenzy for anything AI, reaching the top of that hype curve. This is also reflected in the huge level of investment in AI and its supporting infrastructure. Let’s quickly review some mind-boggling stats about the investments made.

/ Gartner predicts a spend of 2.5 trillion dollars in 2026 after record spending of 1.7 trillion in 2025.²

/ In 2025, global data centre demand was at 82 gigawatts, of which 44 gigawatts were used for AI workloads. This is predicted to rise to 219 gigawatts (of which 156 will be for AI workloads) by 2030.³

/ We also reached some really mind-boggling levels of spend on individual AI projects. Here's a couple of well known examples:

‣ OpenAI’s flagship Stargate project targets 10 gigawatts of total AI infrastructure capacity by 2029 with $500 billion of capital being deployed over four years. This initiative involves Open AI, Softbank, Oracle, MGX and Nvidia.

‣ Meta is building a huge data centre called Hyperion which will supply their AI lab with five gigawatts of computational power. The lab’s footprint will be big enough to cover the equivalent of most of Manhattan. Meta expects to invest hundreds of billions into AI over the next several years.

/ Nvidia – the company behind the GPUs and specialist gear that make up the backbone of AI compute had a 114% increase in revenue from the previous year, up by $130.5 billion. The leather jacket wearing CEO Jensen Huang stated Nvidia has a $500 billion dollar pipeline of orders for Blackwell and Rubin GPUs through 2025 and 2026.

/ Memory such as DDR5 DRAM and NAND flash become greatly more expensive as AI data centers buy up all the capacity. This has greatly affected the average consumer and is set to continue as data centers are predicted to consume 70% of memory chips made in 2026.

Source Wikipedia⁴

On a side note: these stats you see on capacity demand in gigawatts are based on maximum instantaneous power draw at any moment. Your average nuclear reactor according to the US department of energy produces 1 gigawatt of power output - ⁵ so it really puts into perspective how huge these investments are. 

Where are we now?

Gartner itself stated in July last year that we would be entering the ‘trough’ of disillusionment phase as the year progressed.⁶ It’s perhaps been hard to perceive that as for the entire year we were hit with headline after headline about massive AI investments. Slowly however the cracks have been forming, and in some places, these have started to become giant crevices.

These issues are made up of two problems. The first issue is around practical application of AI. This is a very large topic and warrants an article/blog all unto itself (watch this space). In a nutshell however - it boils down to there being a whole gradient in terms of successful application. As it turns out you can’t just whack every problem with an AI hammer and expect a great outcome. The now famous (or infamous?) MIT Study - “The Gen AI Divide - State of AI in Business in 2025” stated that “only 5% of integrated AI pilots are extracting millions in value, whilst the majority remain stuck with no measurable P&L impact".⁷ There are however plenty of examples of excellent applications of AI that save a lot of time, resource and even lives in some cases. These successful use cases tend to be bespoke tailored to more niche uses cases that can leverage large quantities of high-quality data within low context window tasks.

The second problem is arguably much larger but stems from the two main issues mentioned in this post so far. The first is the enormous investments in AI infrastructure - a large portion of which is built off leverage, borrowing and future commitments. The second is the actual lack of value and revenue generated as of now. Right now, there is lots of spend, but just not much revenue. In 2025 Alphabet, Meta, Microsoft and Amazon spent more than $380 billion dollars of infrastructure roll out.⁸ However, all the large hyperscalers are being rather quiet when it comes to explicitly disclosing how much money they are actually making out of AI specific compute:

/ Microsoft was the first (and last) major hyperscaler to explicitly quantify its AI business as a standalone metric. In their Q2 2025 earnings call, CEO Satya Nadella stated an annual run rate of $13 billions of AI specific revenue. Since then, no explicit numbers have been disclosed.⁹

/ In their April 2025 shareholder letter, Amazon CEO Andy Jassy stated that "AI-specific revenue now represents 'a multi-billion-dollar annual revenue run rate'.¹⁰

/ Alphabet hasn’t disclosed their AI explicit revenue figures. The best we have is a statement from CEO Sundar Pichai in his Q3 2025 earnings call that “over 70% of Google Cloud customers now use at least one of its 13 AI product lines, each generating over $1 billion in annual recurring revenue”.

/ Meta have disclosed no revenue numbers at all, but reality labs, their most AI forward operating segment reported a $4.4 billion operating loss on $470 million of revenue.¹¹ 

/ OpenAI have very recently stated that they expect to meet their annual revenue target of $13 billion¹²  and anthropic have approached an annualised run rate of $7 billion.¹³ 

/ OpenAI burnt $8.5 billion in cash and expect to greatly increase that number to $115 billion through 2029.¹⁴

This continuing build out will require roughly $3 trillion dollars of funding through to 2028. $1.4 trillion of this is covered by the cashflow of the big US tech companies, but there is a gap of $1.5 trillion that needs to be covered from other sources such as private credit.¹⁵

I’m no financial analyst, but I think it’s fair to say as it stands that the mind-boggling investments of trillions of dollars are not going to pay off for a very long time if those cash and revenue numbers don’t drastically improve. Investors and private credit will eventually want to see some kind of payback. As AI apps and tools continue to show mixed results, this is becoming an increasingly bleak prospect.

Let’s be clear - I do believe artificial intelligence can bring major value to businesses and individuals for specific use cases - but they do not justify the current spend levels. Whilst large, diversified companies like Google, Microsoft and Amazon are well diversified should this prove to be an actual bubble, OpenAI is much closer to the cliff edge. 

The Fallback plan (or the plan all along?)

To save face - it’s clear that OpenAI looks at advertising as a revenue booster. In a similar manner, Netflix introduced an ad supported tier in November 2022 in a bid to ensure a more consistent revenue stream. This has since proved very successful – driving a predicted $3 billion dollars of revenue¹⁶ in 2025 against total revenues of 11.5 billion. It’s a now common playbook that tech businesses have monetised once adoption levels have increased i.e. Amazon, Spotify etc. OpenAI is now at those adoption levels (700 million users) and has the potential to monetised ads at a whole different level due to its increasing adoption as a search tool and its rich set of data it has on users.

ChatGPT is starting to get closer to Google’s numbers in terms of raw queries - 2.5 billion a day versus 8.5 billion for Google.¹⁷ Furthermore the prize is potentially far greater than companies like Netflix or Spotify can reach - Google’s ad revenue for Q4 alone was $72.5 billion.¹⁸ In September 2025, an OpenAI paper showed just how much of their messages were used for ‘seeking information’ i.e. 'search type' queries:

The 24.4 % represents a 10% increase from the previous year - something I’m sure google will be keeping a very close eye on. Considering the higher level of interaction that ChatGPT provides over Google via its ‘conversation’ mechanism, ChatGPT will also potentially have much more intimate profiling of its users.

Here come the Ads

So far OpenAI has announced that they will begin testing Ads for Go and Free tiers in the US. Plus, Pro and Business accounts are not affected by this test (i.e. B2C only). We know that ad personalisation will also be optional, but the fact that this exists means they will be harvesting user data for profiling. 

There is no information so far on exactly how this will work from a technical perspective, but I’d guess it will be similar to Facebook’s mechanism for audience selection, budget and bidding. OpenAI states that advertising will be a richer experience in their platform, because users will be able to ask questions about the advertised products. They also state that it will specifically be good for small businesses and level the playing field allowing users to ‘discover options they might never have found otherwise’. There is clearly also an appetite for it - Adobe found that as of last year¹⁹ a sample of 1000 people in the US, 77% use ChatGPT as a search engine and that 36% of those discovered a new product or brand via ChatGPT and two in three marketers are keen to increase AI visibility. 

For now, the vast majority of digital marketers will not be able to access OpenAI as a marketing channel. Most likely, a few select US businesses will be ‘invited’ to join the beta program. In the meantime however - you can take several steps that are applicable to almost anyone in our industry - but especially those that might want to advertise via a future OpenAI channel, without knowing the exact detail of what’s coming:

/ Work on your single customer view. If you can unify and profile a patchwork of known and unknown data into a single record - it will be immensely valuable when matching to audiences and understanding conversion

/ Run workshops and strategy sessions on how you want your brand to be used within ChatGPT i.e. how can we be discovered, compared and cross/upsold, how will a user most likely show intent for our products or services

/ What would segments look like for high interaction conversational environments i.e. researching, category explorers

/ Ensure those segments are possible to activate in ad platforms

/ Consider types of conversations that you don’t want to be activated or matched to

Written by Andrew Thornhill, VP of Data and Technical Services at Shaw/Scott Europe

Need help with Martech, Data and Digital Marketing Strategy? Get in touch at inspire@shawscott.com

Sources:

1. Stanford University Human Centered Artificial Intelligence. (2025). 2025 AI Index Report. https://hai.stanford.edu/ai-index/2025-ai-index-report/public-opinion

2. Gartner. (January 15, 2026).Gartner Says Worldwide AI Spending Will Total $2.5 Trillion in 2026. https://www.gartner.com/en/newsroom/press-releases/2026-1-15-gartner-says-worldwide-ai-spending-will-total-2-point-5-trillion-dollars-in-2026

3. McKinsey & Company. (May 20, 2025). Data center demands. https://www.mckinsey.com/featured-insights/week-in-charts/data-center-demands

4. Wikipedia. 2024–2026 global memory supply shortage https://en.wikipedia.org/wiki/2024%E2%80%932026_global_memory_supply_shortage

5. US Department of Energy - Office of Nuclear Energy. (March 31, 2021). How Much Power Does A Nuclear Reactor Produce? https://www.energy.gov/ne/articles/infographic-how-much-power-does-nuclear-reactor-produce

6. Haritha Khandabattu, Gartner. (July 8, 2025).The 2025 Hype Cycle for Artificial Intelligence Goes Beyond GenAI. https://www.gartner.com/en/articles/hype-cycle-for-artificial-intelligence

7. Mit Nanda, Challapally et al, The Gen AI Divide. (May 20, 2025). State of AI in Business in 2025. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf

8. Ari Levy, CNBC. (October 31, 2025).Tech’s $380 billion splurge: This quarter’s winners and losers of the AI spending boom. https://www.cnbc.com/2025/10/31/tech-ai-google-meta-amazon-microsoft-spend.html

9. Microsoft Investor Relations. (January 29, 2025). Earnings Release FY25 Q2. https://www.microsoft.com/en-us/investor/earnings/fy-2025-q2/press-release-webcast

10. Amber Jackson, Data Center Magazine. (April 15, 2025). AWS Revenue Hits US$108bn as Amazon Accelerates AI Push. https://datacentremagazine.com/technology-and-ai/amazon-bets-big-on-ai-as-aws-revenue-reaches-108bn

11. Jacob Falkencrone, Saxo Bank. (October 30, 2025). Meta earnings: record revenue overshadowed by soaring AI costs. https://www.home.saxo/content/articles/equities/meta-earnings-30102025

12. Khac Phu Nguyen, Yahoo Finance. (December 22, 2025). OpenAI Hits 70% Compute Margins as It Pushes Paid ChatGPT Amid AI Race. https://finance.yahoo.com/news/openai-hits-70-compute-margins-154356502.html

13. PM Insights. (November 10, 2025). Anthropic Approaches $7B Run Rate in 2025, Outpaces OpenAI. https://www.pminsights.com/insights/anthropic-approaches-7b-run-rate-in-2025-outpaces-openai

14. Sri Muppidi, The Information. (September 5, 2025). OpenAI Says Its Business Will Burn $115 Billion Through 2029. https://www.theinformation.com/articles/openai-says-business-will-burn-115-billion-2029

15. Dan Milmo, The Guardian. (November 2, 2025). Boom or bubble? Inside the $3tn AI datacentre spending spree. https://www.theguardian.com/technology/2025/nov/02/global-datacentre-boom-investment-debt

16. Colin Dixon, N Screen Media. (October 7, 2025). Netflix’s $3 Billion Ad Boom: A Success or Just Half the Story? https://nscreenmedia.com/netflixs-3b-ad-boom-a-success-or-half-the-story/

17. Aaron Chatterji, Cunningham et al, National Bureau of Economic Research. (September 2025). How People Use Chat GPT. https://www.nber.org/system/files/working_papers/w34255/w34255.pdf Torbjørn Flensted, SEO.AI. (December 2, 2024). How Many People Use Google? Statistics & Facts (2025) https://seo.ai/blog/how-many-people-use-google

18. Jeremy Goldman, E Marketer. (February 5, 2025). Google posts $96.5 billion in Q4 revenue, but ad growth is only half of Meta’s. https://www.emarketer.com/content/google-posts--96-5-billion-q4-revenue--ad-growth-only-half-of-meta-s

19. Adobe Express. (July 7, 2025). How ChatGPT is changing the way we search. https://www.adobe.com/express/learn/blog/chatgpt-as-a-search-engine