The AI race has its price, and business customers will pay part of the bill
Amazon and Alphabet are spending more cash on AI infrastructure than they earn, and the stock market has clearly punished them for it. The costs of this race will also reach companies: through cloud, hardware, and memory prices.
On Thursday, Alphabet lost about $499.8 billion in market capitalization in a single session, and its shares fell by more than 12%. According to data cited by “The Wall Street Journal,” this was the second-largest one-day drop in a company’s value in the history of the U.S. stock market. That same evening, Amazon shares were down nearly 7% in after-hours trading. Both companies are now spending more on data centers and AI chips than their current operations bring in.
For a manager putting together an IT budget for 2027, this is more than a stock market curiosity. Servers bought for hundreds of billions of dollars must pay for themselves eventually. Part of that return will be funded by cloud customers, companies buying hardware, and those paying for AI services.
Amazon and Alphabet are spending more than they earn
In Q3 2026, Amazon spent $77.3 billion on capital expenditures (CapEx, meaning purchases of servers, buildings, and infrastructure). A year earlier, it was $35.1 billion. Free cash flow, meaning the cash left after paying for investments, amounted to negative $42.4 billion on a 12-month basis. A year earlier, it was positive $14.8 billion. The company maintained its plan to spend about $220 billion in all of 2026. Some industry analyses cite about $200 billion, which may result from different announcement timings.
The market also punished Amazon for its guidance. The company expects Q4 revenue of $240 billion to $247 billion. The midpoint of that range, $243.5 billion, comes in below the analyst consensus ($244.51 billion). Sales in Q3 rose 12%, to $201.9 billion, but did not reach the expected $202.2 billion. In practice, Amazon’s annual investments will be close to its entire holiday-season sales.
Alphabet had negative free cash flow of $9.1 billion in Q3. Advertising revenue came in at about $82 billion, below the expected $83.4 billion. Google’s chief business officer Philipp Schindler admitted that advertising performance was affected by the growing popularity of AI Mode, the chatbot conversation mode in the search engine. According to Search Engine Journal, the number of queries is growing at its fastest pace in five years, but each query generates less advertising revenue. CFO Anat Ashkenazi announced that capital expenditures in 2027 will increase significantly compared with this year’s forecast of $195-205 billion.
The cloud is growing, but returns are not evenly distributed
The giants have strong arguments. Revenue from AWS, Amazon’s cloud, rose 39% to $45.98 billion, and the segment’s operating profit jumped from $11.4 billion to $17.8 billion. Google Cloud grew 103% to $30.71 billion, and its order backlog reached $561 billion.
Our investments in AWS features and infrastructure are delivering results. In the third quarter, we recorded more than 39% year-over-year revenue growth, and projected revenue growth in 2027 and 2028 is expected to exceed 40%
On the other side are the skeptics. Ed Zitron, an independent commentator and critic of the AI industry, estimates that Microsoft, Meta, Amazon, and Alphabet spent about $560 billion on AI infrastructure over two years and earned about $35 billion from it. That is his calculation, not company data. Torsten Slok, chief economist at Apollo Global Management, points out that outside the largest tech companies in the S&P 500, AI investments are not clearly translating into either revenue per employee or margins. Piotr Miliński of Quercus TFI indicates that demand and monetization are real, but warns that this does not mean every investment is profitable.
Our conclusion: for now, those who sell computing power and components are making good money from AI. The rest of the economy is only just checking whether these purchases result in real savings.
Memory and hardware are getting more expensive already
The costs of the race are most visible in hardware. SK Hynix, a Korean memory manufacturer and one of Nvidia’s main suppliers, said that its HBM and DRAM production is practically sold out through 2029. HBM is high-bandwidth memory used in AI accelerators, while DRAM is the standard working memory of computers and servers. Long-term contracts cover more than half of the company’s capacity for 2027-2028. Management believes shortages may last until 2030, and that 2027 could be the hardest year for supply in the industry’s history. New facilities will not help quickly: the production hall at the Y1 plant in Yongin is set to start in February 2027, and the Y2 plant not before June 2029.
The effects are already visible in price lists. A 32 GB DDR5-6000 memory kit rose in price over a year from about $72 to $392, and laptop maker Framework raised DDR5 module prices by 50%. Gartner forecasts that because of more expensive RAM, global PC shipments will fall by 10.4% in 2026, and smartphones by 8.4%. Intel reports the same divergence: sales of processors for data centers and AI rose 78% to $7.3 billion, while the personal computer segment shrank by 5%. According to the company, weakness in the consumer market will last at least until mid-2027.
How the costs of the race may reach companies
For now, there have been no announced sharp increases in the basic price lists for cloud computing power. Cost pressure reaches customers indirectly: through more expensive servers and computers, additional cloud fees (data transfer, API calls, storage), and the prices of AI services and subscriptions. Google’s move to expand ads in AI Mode to more markets shows that companies under investor pressure are looking for new sources of revenue. More may be added to business customers’ invoices.
The second cost is dependence. Computing power is sold by a handful of hyperscalers (the largest cloud providers, such as Amazon, Microsoft, and Google), and memory by a few manufacturers that already have their capacity allocated for years. A small or medium-sized company has little bargaining power in these negotiations.
What to include in the IT budget for 2027
When putting together a realistic spending plan, a few steps are helpful:
Alphabet has already announced that it will spend even more on investments in 2027. A company that calculates today how much AI really gives it and how much its dependence on suppliers costs will find it easier to defend its budget when those expenses start coming back in price lists.