US Treasury yields have climbed toward levels not seen in nearly two decades. Alongside federal deficits and expensive oil, a new pressure point has emerged: the surge in borrowing by technology companies financing data centres and artificial intelligence infrastructure.
Artificial intelligence has been one of the main explanations for optimism about the US economy in recent years, with hundreds of billions of dollars flowing into chips, servers, power networks and data centres. Now the same investment boom is revealing another side: a rapidly growing demand for borrowed capital.
This week, the yield on the 30-year US Treasury bond climbed as high as 5.337%, its highest level since 2007. It fell sharply after intervention by the Treasury Department, but by August 21 had moved back toward 5.25%, showing how fragile the initial relief in the bond market proved to be.
Federal finances remain one of the clearest sources of pressure. US gross federal debt has surpassed $40 trillion, while large budget deficits require Washington to keep issuing new securities. At the same time, high oil prices and the prolonged conflict involving Iran have added to inflation concerns.
But another large borrower has entered the competition for capital: the artificial intelligence industry. Alphabet, Amazon, Meta, Oracle and other technology groups are increasingly financing AI infrastructure not only with internal cash flow, but with major corporate bond sales.
According to Daycom’s analysis, US Treasury yields are not rising simply because AI companies are “taking money away” from the government. The mechanism is broader: massive investment increases corporate bond supply, supports economic growth and strengthens expectations that interest rates may stay elevated for longer.
Only a few years ago, the largest technology groups could fund most of their data-centre expansion directly from operating cash flow. In 2024, major hyperscalers issued roughly $20 billion of investment-grade debt. By 2025, that figure had risen above $130 billion.
The scale has increased again in 2026. Bond issuance by Amazon, Alphabet, Meta and Oracle has approached $200 billion since the start of the year, making artificial intelligence one of the most important drivers of new investment-grade corporate debt.
This does not mean the companies are in financial distress. Most remain highly profitable and carry strong credit ratings. The issue is that demand for computing capacity is growing so quickly that even enormous internal cash generation is no longer the only practical way to fund the expansion.
Alphabet, for example, raised the equivalent of almost $3.9 billion in Australia’s bond market in August. Investor demand far exceeded the size of the offering, but the deal illustrates how aggressively large technology companies are diversifying funding sources across global debt markets.
The next phase could be larger still. Broadcom has been discussing tens of billions of dollars in borrowing tied to AI infrastructure, with some financing structures potentially approaching $100 billion in total size.
Vanguard has estimated that roughly a quarter of record investment-grade corporate bond issuance in the first half of 2026 was already connected to artificial intelligence spending. The largest hyperscalers alone accounted for a meaningful share of the market.
For bond investors, this marks an important structural change. A portfolio manager seeking high-quality long-term debt once chose between Treasuries and a relatively limited supply of blue-chip corporate bonds. Now technology giants are bringing tens of billions of dollars of new securities to market on a regular basis.
To absorb that volume, investors need to be compensated. In bond markets, that means higher yields. If new corporate bonds offer more attractive returns, capital can shift away from existing securities, pushing their prices lower and market yields higher.
That change can already be seen in corporate spreads. Alphabet sold 10-year debt in the spring at roughly 0.63 percentage point over Treasuries, while a later issue required a spread of about 0.85 point. Amazon’s comparable spread rose from roughly 0.55 to about 0.8 point.
Oracle, which has a weaker credit profile than some of the largest AI companies, has faced an even sharper repricing. The premium on its 10-year debt rose from around 1.05 percentage points to about 1.45 points over comparable US government bonds.
The concern is not that these companies are suddenly likely to default. Investors are reacting to the sheer volume of supply. When new bonds keep arriving with more attractive pricing, securities purchased earlier can quickly look expensive by comparison.
Still, the direct shift of money from Treasuries into corporate debt explains only part of the move. The US government bond market is so large that even hundreds of billions of dollars in AI-related issuance remain relatively modest in comparison.
The more important channel may be macroeconomic. Building data centres requires land, steel, transformers, gas turbines, semiconductors, power infrastructure and large workforces. That spending supports demand even when high interest rates would normally be expected to cool the economy.
This, in turn, affects expectations for the Federal Reserve. If AI investment keeps growth stronger while inflation remains persistent, the Fed has less reason to cut rates quickly. Long-dated bonds then need to offer higher yields to compensate investors.
That is what turns the AI boom into a policy paradox. It may eventually raise productivity and the economy’s long-term output, but before those gains fully arrive it creates an enormous present-day demand for capital, electricity, equipment and construction capacity.
The latest rise in yields, however, cannot be attributed to artificial intelligence alone. The 30-year rate climbed toward its highest level since 2007 amid expensive oil, geopolitical risk, inflation concerns and record federal borrowing.
AI matters because it adds another structural pressure to a market already absorbing an extraordinary amount of debt. The US government is borrowing to finance deficits, corporations are borrowing for data centres and energy infrastructure, and investors are demanding compensation for inflation and the risk of further bond-price declines.
The Treasury Department’s response showed how seriously Washington viewed the rise in long-term rates. On August 19, it increased the size of some buyback operations in longer-dated government securities in an effort to support liquidity and demand.
The mechanics are straightforward. When the Treasury buys back older bonds, it creates additional demand, supports prices and pushes yields lower. Following the announcement, the 30-year yield briefly dropped by close to 10 basis points.
But this is not the same as quantitative easing by the Federal Reserve. The Treasury is managing the structure and liquidity of its own debt rather than creating new banking reserves, and the size of the operations remains small relative to a Treasury market worth tens of trillions of dollars.
The effect therefore proved short-lived. By August 20 and 21, yields were climbing again, with the 30-year rate returning to around 5.25%. Investors were effectively signalling that technical buybacks do not remove the fundamental reasons capital has become more expensive.
That creates a political problem for the Trump administration. The White House has promised lower borrowing costs for American households, yet long-term Treasury yields directly influence mortgage rates, corporate borrowing and other forms of credit.
In August, the average 30-year mortgage rate again approached roughly 6.7%. For consumers, that means the AI investment boom can be felt far beyond Silicon Valley — in home affordability and monthly borrowing costs.
Artificial intelligence is therefore helping and complicating the US economy at the same time. It supports investment, employment and expectations of higher productivity, but may also contribute to keeping the price of money elevated for households, businesses, large corporations and the federal government.
For the technology companies themselves, current borrowing costs are not yet a major threat. Their balance sheets remain strong and profits are large enough to service the debt. The bigger question is whether the enormous investment in data centres will generate returns sufficient to justify still more capital spending.
If AI rapidly lifts productivity across the economy, today’s debt burden may ultimately look like the cost of future growth. If monetisation lags behind spending, investors could begin demanding even larger premiums from companies that until recently could borrow at rates close to those of the government.
For the Treasury market, the challenge is broader. It must finance record federal debt, absorb geopolitical inflation shocks and compete with an unprecedented stream of high-quality corporate bonds. Artificial intelligence did not create that problem, but it has made it considerably more complex.
The rise in US Treasury yields should therefore not be read as a market verdict against artificial intelligence. It is better understood as a repricing of global capital itself, as AI shifts from a software story into an infrastructure build-out large enough to affect the entire financial system.
If the era in which technology giants could build the future almost entirely from their own cash really is ending, the consequences will stretch far beyond the technology sector. Mortgage rates, corporate borrowing costs and Washington’s debt-service bill are increasingly tied to the same competition — who gets the next trillion dollars of capital, and at what price.
