AI Industry | Hidden austerity measures and cost-cutting measures in generative AI – including shrinking word count: Less is cheaper – Image: Xpert.Digital
The hidden cost crisis among providers of generative artificial intelligence
The great AI illusion: Billions in losses and austerity measures – what the tech giants are hiding from you
Artificial intelligence is experiencing an unprecedented boom, but behind the shiny facade of the technological revolution lie massive financial challenges. What users perceive as minor technical adjustments turn out, upon closer inspection, to be desperate attempts by major AI providers to get their exploding costs under control. Google and OpenAI made changes almost simultaneously in September 2024. At first glance, these changes appear unrelated, but they reveal a larger problem facing the entire industry: the monetization of AI services is dramatically lagging behind the massive investments and operating costs.
Generative AI (Generative Artificial Intelligence) is a branch of artificial intelligence that specializes in generating new content rather than just analyzing or classifying existing data.
This development is particularly noteworthy because it demonstrates how even multi-billion dollar tech giants are suffering from the financial pressure of their own innovation. While OpenAI, despite a $157 billion valuation, forecasts losses of $5 billion for 2024, Google is quietly implementing measures that drastically increase the costs of data collection and processing. These seemingly small changes have far-reaching consequences for the entire digital landscape and point to a larger structural crisis in the AI industry.
Google's silent revolution
Disabling the num=100 parameter
On September 14, 2024, Google made one of the most significant changes to its search infrastructure in years without much publicity. The URL parameter num=100, which had enabled the simultaneous display of 100 search results for decades, was completely disabled. This technical innovation had far-reaching consequences for the entire SEO industry and AI applications.
While the parameter was a convenience feature for regular users, it was of vital importance to the SEO industry. Virtually all major SEO providers, such as Ahrefs, Sistrix, SEMrush, and specialized ranking tools, used this parameter for efficient data collection. With a single call, they could capture the complete top 100 rankings, which was both more cost-effective and faster than retrieving results page by page.
The deactivation of this parameter led to a cost explosion in the SEO industry. Ranking tools now have to perform ten separate queries to obtain the same amount of data that was previously possible with a single query. This represents a tenfold increase in query costs and has already posed serious technical and financial challenges for several tool providers.
Impact on website operators
The effects were immediately apparent in Google Search Console: 87.7 percent of all websites examined experienced a drastic decline in measured impressions. At the same time, paradoxically, the average position of the websites improved, as fewer "low-ranking impressions" were recorded from positions 11 to 100.
Many publishers are reporting dramatic traffic declines. News sites like derwesten.de lost 51 percent of their search traffic, express.de 35 percent, and focus.de 33 percent. The reasons are complex: In addition to technical changes at Google, AI reviews are also playing a role, causing users to click less on external websites.
Motivation behind Google's strategy
The reasons for Google's decision are complex. On the one hand, the change significantly reduces server load, as less data needs to be processed per request. On the other hand, it makes it more difficult for bots and scraping services to collect data on a massive scale, which is particularly relevant given the increased interest of AI companies in Google's data.
Another aspect is user experience: Google wants to encourage users to return to traditional search, where they click through multiple pages and spend more time in the Google environment. This strengthens Google's position as a central point of contact for information and makes the company less dependent on redirecting users to external websites.
Suitable for:
OpenAI's desperate cost control: The dramatic decline in citations
Parallel to Google's change, OpenAI made an equally significant change to ChatGPT. Since September 11, 2024, the AI chatbot has displayed significantly fewer citations and links to external websites. This reduction affected all industries and content types equally, regardless of the quality of the content or the domain authority of the sources.
The numbers are dramatic: According to analyses, citations in ChatGPT have dropped by up to 90 percent. Free ChatGPT accounts are particularly affected, while users of the paid versions experience significantly fewer restrictions. This development exacerbates the already existing discrepancy between the traffic websites receive from Google and the significantly lower traffic from ChatGPT.
The explosive increase in costs for web search features
OpenAI is under enormous financial pressure. The company expects losses of around $5 billion by 2024, while operating ChatGPT costs up to $700,000 daily. Reducing web searches and citations is an obvious cost-cutting measure, as each web search requires additional computing resources and API calls.
The cost of OpenAI's web search feature has increased significantly. While earlier models offered free access to web searches, newer models charge full price for search tokens. An example illustrates the cost dilemma: A query with GPT-4o cost $0.13, while the same query with GPT-5, with the more extensive web search tokens, cost $74.
For its current web-search models, OpenAI charges $25 per 1,000 views for gpt-4o and gpt-4.1, while more powerful models like GPT-5 and the o-series cost as little as $10 per 1,000 views. This pricing clearly demonstrates why OpenAI is making drastic cuts in the delivery of web-based information.
The shrinking word count: Less is cheaper
In addition to the reduction in source citations, users have noticed another, more subtle change: ChatGPT's responses have become noticeably shorter and less detailed. What at first glance may seem like an optimization toward greater conciseness is actually another effective cost-saving measure. Every generated word—or, more technically precise, every token—consumes computing power and thus incurs direct costs. By systematically shortening responses, OpenAI reduces the operating costs per query.
This trend is particularly evident in direct comparisons with competitors like Anthropic's Claude or Google's Gemini. These models often continue to provide more detailed, nuanced, and in-depth answers to the same queries. While some competitors still rely on richness of detail as a quality indicator, OpenAI appears to be deliberately reducing its word count to cope with the financial burden of its immense user base.
For users, this means additional work. Instead of a comprehensive answer, they often receive only a superficial summary and must extract the desired depth of information themselves through targeted queries (so-called "prompt chaining"). Each of these follow-up queries represents a new transaction, which, while cheaper individually, costs the user more time and effort. This measure fits seamlessly into the strategy of reducing citations: Both are degradations of the user experience that, when combined, lead to significant cost savings and are intended to offset the company's massive financial deficit.
Shorter answers, fewer sources: Do you also notice how ChatGPT is secretly cutting costs?
The loss-making business with premium subscriptions
Particularly problematic is that even the more expensive ChatGPT Pro subscription, costing $200 per month, is generating losses as users use more services than expected. CEO Sam Altman called this situation "crazy," acknowledging the challenges of covering costs.
OpenAI CEO Sam Altman admitted that the company is currently losing money on the $200 subscription: "People are using it much more than we expected." This surprising finding demonstrates how difficult it is for AI companies to predict the true cost of use and calculate appropriate pricing.
The surprising connection between both changes
The temporal proximity of both events is more than a coincidence. ChatGPT frequently relies on current web information for its responses, accessing Google results directly or indirectly via scraping services. Disabling the num=100 parameter also significantly complicates the efficient collection of web data for ChatGPT and other AI systems.
AI applications like ChatGPT rely on extensive, up-to-date web data to generate relevant and accurate answers. The num=100 parameter enabled these systems to quickly and cost-effectively capture large volumes of search results and select the best sources for their answers.
By disabling this parameter, AI systems now have to perform significantly more individual queries, which increases costs exponentially. This explains why OpenAI simultaneously reduced the citation frequency—the costs of providing up-to-date web information were simply no longer economically viable.
The $800 billion funding gap
Bain & Company's alarming forecast
A recent study by Bain & Company reveals a threatening funding gap in the AI industry. By 2030, AI companies like OpenAI, Google, and DeepSeek will need to generate approximately $2 trillion annually to cover the rising costs of computing power and infrastructure. However, the consultants expect the industry to fall about $800 billion short of this target.
David Crawford, Chairman of Bain & Company's Global Technology Practice, warns urgently: "If current scaling laws prevail, artificial intelligence will increasingly burden global supply chains." This discrepancy between required and expected revenues raises fundamental questions about the valuation and business models of the AI industry.
Massive investments versus unclear returns
Major US technology companies are driving their AI investments to unprecedented levels. Microsoft, Meta, and Google are planning a combined $215 billion for AI projects by 2025. Amazon has announced an additional $100 billion in investments. These expenditures will primarily go toward expanding data centers and developing new AI models.
Investments have more than doubled since the launch of ChatGPT. By 2024, the four largest tech companies had invested a combined $246 billion in AI—a 63 percent increase over the previous year. By the early 2030s, annual spending on AI could exceed $500 billion.
Energy demand and infrastructure challenges
Bain & Company predicts that additional global demand for computing power could climb to 200 gigawatts by 2030, half of which will be in the United States. AI data center power consumption will increase from 50 billion kilowatt-hours in 2023 to approximately 550 billion kilowatt-hours in 2030—an elevenfold increase.
This massive expansion will have significant environmental impacts. Despite the expansion of renewable energy, greenhouse gas emissions from data centers will rise from 212 million tons in 2023 to 355 million tons in 2030. Water consumption for cooling will nearly quadruple to 664 billion liters over the same period.
Suitable for:
The DeepSeek shock as a turning point
Cost-effective innovation from China
The Chinese startup DeepSeek has shaken up the AI industry with its R1 model. With an estimated development cost of just $5.6 million, the company has developed a model that can compete with the significantly more expensive US models. By comparison, OpenAI's GPT-4o cost approximately $80 million to develop.
DeepSeek's pricing massively undercuts the competition. The company's models are 20 to 40 times cheaper than corresponding models from OpenAI. DeepSeek's Reasoner model costs 53 cents per million input tokens, while OpenAI's o1 model costs $15 for the same number.
Impact on industry dynamics
DeepSeek's success challenges previous assumptions in the AI industry. The company proves that cutting-edge AI is possible even without billion-dollar budgets, putting established providers under considerable pricing pressure. This development highlights an interesting side effect of the US export restrictions: The technical limitations forced the company to engage in software innovations to make optimal use of the available hardware.
Within a few weeks, DeepSeek's AI assistant captured 21 percent of the global LLM user share and displaced ChatGPT as the most popular free app in Apple's App Store. This rapid market penetration highlights the volatility of the AI market and the danger for established providers with cost-intensive business models.
Our recommendation: 🌍 Limitless reach 🔗 Networked 🌐 Multilingual 💪 Strong sales: 💡 Authentic with strategy 🚀 Innovation meets 🧠 Intuition
From local to global: SMEs conquer the global market with clever strategies - Image: Xpert.Digital
At a time when a company's digital presence determines its success, the challenge is how to make this presence authentic, individual and far-reaching. Xpert.Digital offers an innovative solution that positions itself as an intersection between an industry hub, a blog and a brand ambassador. It combines the advantages of communication and sales channels in a single platform and enables publication in 18 different languages. The cooperation with partner portals and the possibility of publishing articles on Google News and a press distribution list with around 8,000 journalists and readers maximize the reach and visibility of the content. This represents an essential factor in external sales & marketing (SMarketing).
More about it here:
AI creates winners and losers: Who will survive the new internet order?
The dramatic impact on website traffic
The decline of AI Overviews
A study by Authoritas shows a significant drop in click-through rates for publisher websites in the UK, by about half, due to AI Overviews. According to the study, the click-through rate drops by 47.5 percent on desktop and 37.7 percent on mobile in the presence of AI Overviews. Even when a website ranks at the top of Google AI Overviews, there is only a minimal improvement in clicks.
A new study by SEO expert Kevin Indig and UX researcher Eric Van Buskirk systematically examines the use of Google's AI Overviews. In a complex setting, 70 users were observed performing eight realistic search tasks. This resulted in approximately 400 AIO interactions, providing insights into how significantly web search behavior changes under the influence of AI.
AI overviews then significantly reduce the click-through rate on external links. On desktop, the click-through rate drops by up to two-thirds, and on mobile devices by almost half. Users increasingly rely on the information in the AI overview – especially for simple or standardized queries.
Suitable for:
The collapse of traditional web traffic
Since Google integrated generative AI answers into search in the US in May 2024 and in Germany at the end of March 2025, content providers such as news sites, blogs, and forums have been concerned about a significant decline in their visitor numbers. If the answer is already on the search page, people often don't click on the original source.
Publisher Mail Online's CTR dropped by over 56 percent due to AI Overviews. Some websites experienced a drop in organic traffic of 18-64 percent. Organic CTR can drop by as much as 70 percent in the presence of AI.
Built In reports that organic search traffic for publishers could drop by 25 percent by 2026. Even top-three positions suffer massive CTR drops due to AIOs; positions 4-10 see declines of up to 50 percent. A study of 1,000 SMB websites found that 68 percent experienced significant organic traffic losses after implementing AI search features.
The concentration on a few sources
Referral traffic from ChatGPT to websites has fallen by 52 percent since July 2024. Reddit citations have increased by 87 percent and now account for over 10 percent of all ChatGPT citations. Wikipedia jumped 62 percent from its July low, capturing nearly 13 percent of the citation share.
The top three websites—Wikipedia, Reddit, and TechRadar—accounted for 22 percent of all citations, an increase of 53 percent in just one month. ChatGPT now favors a handful of "answer-first" sources, while brand-related websites lose visibility and miss out on millions of potential referral clicks.
Breakdowns under time pressure
Google's image generation problems
The current problems with Google Gemini manifest themselves on various levels. Users have been reporting fundamental malfunctions in the image technology for weeks, particularly when generating images in the desired formats. The widespread problem primarily affects the creation of images in 16:9 format, which used to be possible without any problems, but is now no longer supported.
Even more serious is the phenomenon where images are supposedly generated but cannot be displayed. Users receive confirmations that images have been successfully created, but only see empty spaces or error messages. This issue occurs in both the web version and the mobile app, rendering the image generation function virtually unusable.
Communication failure and lack of transparency
Google's handling of these apparent system errors is particularly problematic. The company doesn't proactively communicate these issues to users, even though they've been present for weeks. Instead, the system continues to claim that all functions are working properly, while actual performance is significantly degraded.
The current difficulties should not be viewed in isolation, but rather are part of a series of problems with Google's AI systems. In February 2024, Google had to completely disable the human representation in Gemini after the system generated historically inaccurate images. German soldiers were depicted with Asian facial features, and Vikings were given dreadlocks – errors that revealed fundamental problems in the training data preparation.
Suitable for:
Structural problems in AI development
The haste of publications
The recurring problems point to systemic weaknesses in Google's AI development. The company appears to be under enormous time pressure to keep pace with competitors like OpenAI, resulting in hastily released products. This "move fast and break things" mentality may work in other areas of technology, but it proves problematic with AI systems, as errors have a more direct impact on the user experience.
The working conditions of the subcontractors responsible for content moderation and system improvement further exacerbate these problems. Reports of time pressure, low wages, and a lack of transparency in the supply chain raise doubts about the quality of manual system optimization.
Fragmentation of the system architecture
The lack of integration between various Google services reveals structural deficiencies. While Google Photos is gaining new AI image processing features, basic image generation in Gemini isn't working properly. This fragmentation indicates insufficient internal coordination and exacerbates the problems for end users.
The economic consequences
Impact on different user groups
The problems described have concrete impacts on various user groups. Content creators and marketing professionals who rely on reliable image generation must resort to alternative solutions. This not only leads to workflow interruptions but also to additional costs for other tools.
The situation is particularly problematic for users of the paid Gemini Pro version. They pay for advanced features but often receive worse performance than promised. Many have already canceled their subscriptions because the promised improvements have not materialized.
The loss of trust in AI providers
The unreliability of the systems is leading to a loss of trust in Google as an AI provider. Users who rely on the accuracy and availability of the services are increasingly turning to alternative providers. This could weaken Google's position in the highly competitive AI market in the long term.
While OpenAI's DALL-E and Anthropic's Claude deliver more consistent results, Google struggles with fundamental functional issues. What's particularly striking is that even free alternatives often perform more reliably than Google's premium offerings.
Parallels to the dotcom bubble
Similar market dynamics
Current developments draw striking parallels to the dot-com bubble at the turn of the millennium. Back then, the internet hype led to extreme valuations and culminated in a spectacular crash. Today, AI companies face similar challenges: astronomical valuations collide with unclear business models, while the gap between investments and actual revenues continues to widen.
The current valuation of the S&P 500 is equivalent to 38 times the earnings of the past ten years. Only during the dot-com bubble was the valuation higher, as strategists at Morgan Stanley point out. Henry Blodget, a former star analyst of the dot-com era, warns of uncanny parallels to the current AI boom.
The bizarre truth behind the US recovery
George Saravelos of Deutsche Bank puts it in a shocking way: "AI machines are literally saving the US economy." His analysis reveals a paradoxical situation – economic growth isn't coming from revolutionary AI applications, but from the mere construction of "AI capacity generation factories."
Particularly controversial: The bank notes that without this technology-related spending, America would already be in recession. A house of cards made up of billions in investments keeps the world's largest economy afloat. This extreme dependence on AI investments poses systemic risks that extend far beyond the technology industry.
Suitable for:
Solutions and future perspectives
Alternative business models
Rising costs are forcing AI providers to develop new business models. OpenAI is testing usage-based pricing, where customers only pay for the computing power they actually use. Batch APIs already offer cost savings of up to 50 percent for non-time-critical requests.
To achieve profitability, OpenAI is reportedly considering raising prices for its various subscription tiers. The company plans drastic price increases: ChatGPT Plus is expected to rise from $20 to $22 per month by the end of 2024, and even reach $44 per month by 2029.
Specialization as a survival strategy
The rising costs of general-purpose AI systems could lead to the development of specialized solutions. Instead of attempting to answer all queries with up-to-date web data, AI systems could be more selective and only resort to costly web searches for specific query types.
This would promote a diversification of the AI market, with different providers developing different specializations. Some could focus on current information, others on deep expertise without internet connectivity.
New cooperation models
New collaboration models are already emerging between AI providers and content creators. Some publishers are negotiating direct licensing agreements with AI companies to receive a fair share of the use of their content. This development could lead to a new ecosystem in which content creators are directly compensated for the use of their content in AI systems.
Recommendations for various stakeholders
For website operators and content creators
Website owners should diversify their strategies and not rely solely on search engine traffic. Building direct relationships with users through newsletters, social media, and other channels will become more important. At the same time, they should improve the quality of their content to be mentioned in AI-generated answers.
Instead of focusing on quantity, quality is becoming more important. Content creators should focus on creating helpful, unique content that offers real value. The days when pure SEO optimization was enough to achieve top 100 positions are over.
For SEO agencies and tools
SEO agencies need to adapt their services and focus more on the top 20 positions, as these generate the majority of real traffic. The era of comprehensive top 100 analyses is coming to an end, which could free up resources for more in-depth optimization.
Tool providers like Semrush and Accuranker are working feverishly to adapt their systems, but the higher costs are inevitably passed on to customers. Many established tools currently display incomplete or incorrect data because their crawling logic was based on the old num=100 parameter.
For AI companies
AI companies face the challenge of developing sustainable business models that work for both them and content creators. Current practices of free content usage are unsustainable in the long term if they undermine the foundation of their data sources.
To regain user trust, companies like Google must make fundamental changes to their approach. First, more transparent communication about system issues and planned maintenance is required. Users have a right to know when features aren't working properly.
Hidden austerity measures and their consequences: The new power of the tech giants
The seemingly small technical changes at Google and ChatGPT mark a fundamental turning point in the digital information landscape. They demonstrate how dependent the entire internet ecosystem is on the decisions of a few large technology companies.
The combined effect of both changes is accelerating the transformation from a link-based to an AI-mediated internet. This development brings with it both opportunities and risks: Users get answers to their questions faster, but the economic foundations of content creation are fundamentally questioned.
The industry is in a phase of reorientation in which new balances must be established between technology providers, content creators, and users. The coming years will show which players can successfully adapt to the changing conditions and which new business models will emerge.
The $800 billion funding gap predicted by Bain & Company could lead to industry consolidation. Only the financially strongest companies would survive, while smaller providers and startups could disappear from the market. Whether the AI bubble develops into a controlled correction or a dramatic crash depends on whether the industry can develop viable business models in a timely manner.
The hidden austerity measures at major AI providers are just the tip of the iceberg of a much larger structural crisis. While the public continues to be fascinated by the possibilities of artificial intelligence, behind the scenes, companies are desperately fighting for their financial survival. The silent revolution is already in full swing – its effects will shape the digital landscape for years to come.
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your national language!
Konrad Wolfenstein
I would be happy to serve you and my team as a personal advisor.
You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital
I'm looking forward to our joint project.
☑️ SME support in strategy, consulting, planning and implementation
☑️ Creation or realignment of the digital strategy and digitalization
☑️ Expansion and optimization of international sales processes
☑️ Global & Digital B2B trading platforms
☑️ Pioneer Business Development / Marketing / PR / Trade Fairs
🎯🎯🎯 Benefit from Xpert.Digital's extensive, fivefold expertise in a comprehensive service package | R&D, XR, PR & SEM
AI & XR 3D Rendering Machine: Fivefold expertise from Xpert.Digital in a comprehensive service package, R&D XR, PR & SEM - Image: Xpert.Digital
Xpert.Digital has in-depth knowledge of various industries. This allows us to develop tailor-made strategies that are tailored precisely to the requirements and challenges of your specific market segment. By continually analyzing market trends and following industry developments, we can act with foresight and offer innovative solutions. Through the combination of experience and knowledge, we generate added value and give our customers a decisive competitive advantage.
More about it here: