Major technology companies are spending record sums on AI infrastructure—$194 billion in the first half of 2026 alone by four firms—in a race to control compute capacity, mirroring the railway bond bubble of the 1870s that preceded a financial panic and long depression.
Microsoft remains the only hyperscaler funding its buildout without debt, while rivals are increasingly issuing bonds at deteriorating terms, raising questions about whether the returns will justify the investment or trigger a downturn.
What happened
Tech companies are raising unprecedented sums for AI infrastructure—$194 billion so far in 2026 by just four firms (Oracle, Meta, Alphabet, Amazon), driven by a belief that whoever controls the most compute wins. Microsoft alone maintains $19.6 billion in free cash flow last quarter and funds CapEx without debt, while rivals increasingly rely on bond issuance; 86% of bonds issued this year already trade at higher yields than at issuance, and cover has fallen to less than 2x from 5x in February.
Why it matters
A financial historian notes that $600 billion is projected to be invested by major tech companies in 2026—equivalent in scale to the $500 million poured into U.S. railway bonds annually during the 1870s boom that preceded the Panic of 1873, a multi-year depression and four-decade deflation. The parallel raises the question of whether today's compute-at-all-costs race will prove sustainable or create conditions for a financial reckoning if the expected returns do not materialize.
What to watch
Google's internal restructuring—the departure of DeepMind CEO Demis Hassabis and Chief Scientist Jeff Dean, among other researchers—signals a pivot away from frontier AI leadership toward monetizing infrastructure through Google Cloud, where revenue growth has reached 82% year-over-year; the company is reportedly renting capacity to Anthropic and selling over 20% of its TPU shipments from 3Q26 to 4Q27 directly to Anthropic, suggesting the real profit may lie in supply rather than model development.
The article opens with the story of Jay Cooke, a Civil War financier who in 1870 undertook to fund the Northern Pacific Railway's expansion from Duluth to Tacoma. Congress had chartered Northern Pacific in 1864 and granted it 40 million acres of adjacent land, but the company struggled for six years to secure financing, watching competitors Union Pacific and Central Pacific complete their routes (culminating in the golden spike at Sacramento-Omaha in May 1869). When Northern Pacific approached Cooke, it offered a 12 percent commission on every bond sold and $200 in stock per $1,000 in bonds—a deal Cooke could not refuse after institutional investors rejected his pitches.
Cooke's solution was revolutionary: rather than rely on banks or government loans, he pioneered retail bond sales, employing 1,500 salespeople and funding 1,300 newspapers (through advertising and direct payments) to create a media machine backed by his Civil War reputation. This worked until September 1873, when a crash on the Vienna stock exchange and the demonetization of silver tightened global credit; Cooke, who had been funding Northern Pacific from deposits between bond sales, could find no more buyers. His bankruptcy triggered the Panic of 1873, culminating in railroad failures, a multi-year depression, and multi-decade deflation.
The author draws a parallel to the current AI moment via Liaquat Ahamed's book *1873*. Ahamed notes that $500 million invested in U.S. railway bonds annually during the 1870s boom would equal approximately $600 billion in 2026 dollars—roughly what major tech companies are projected to invest in AI infrastructure in 2026. Microsoft CEO Satya Nadella cited the book as essential reading on the company's earnings call. Notably, Microsoft is the only hyperscaler that still generates substantial free cash flow ($19.6 billion last quarter) and funds capital expenditure without debt.
Other hyperscalers are not so disciplined. Between September and November, Oracle, Meta, Alphabet, and Amazon issued $80 billion in combined debt for infrastructure; after raising $108 billion in all of 2025, they have already raised $194 billion in 2026 as of July 7. Bond market stress is evident: 86% of bonds issued this year already trade at higher yields than at issuance, and bond cover (the ratio of demand to supply) has fallen to below 2x from 5x in February. In June, Google announced a $85 billion equity raise, including a $10 billion stake sale to Berkshire Hathaway, signaling its willingness to use "all means at its disposal" to fund supply.
Google's internal dynamics, however, tell a different story. DeepMind CEO Demis Hassabis was technically promoted to chairman but removed from day-to-day operations; Chief Scientist Jeff Dean departed, as did other prominent researchers. SemiAnalysis declared that DeepMind is no longer a frontier lab and that Google has "run out of patience" with Hassabis's vision of world models (multimodal systems) in favor of a more text- and code-centered approach aligned with rival Anthropic. The real winner, SemiAnalysis argues, is Google Cloud, where revenue growth accelerated to 82% year-over-year (from 63% last quarter and 32% a year ago) with margins of 36% (up from 33% and 21% respectively).
Google Cloud's strength stems partly from renting capacity to Anthropic and other large customers. During Google's earnings call, CEO Sundar Pichai explained that the company rents third-party capacity to "very, very large customers" on a short-term basis at "very high" upfront cost, betting that "multiyear" margins will be "highly ROI-positive." The author identifies this customer as almost certainly Anthropic. SemiAnalysis notes that more than 20% of Google's TPU shipments from 3Q26 to 4Q27 are being sold directly to Anthropic, excluding the hundreds of thousands of TPUs Google Cloud already rents to Anthropic. This suggests that rather than dominating frontier AI research, Google is betting that control of infrastructure—the "supply" of compute—will prove more durable and profitable than control of the models themselves.
The article draws a historical parallel between today's artificial intelligence spending spree and the railroad construction boom of the 1870s, when investor Jay Cooke pioneered retail bond sales to finance the Northern Pacific Railway. Cooke's innovation—bypassing institutional lenders by appealing directly to retail investors through media control and patriotic messaging—worked spectacularly until 1873, when credit tightened globally and the resulting collapse triggered a panic and decade-long deflation. The author notes that Microsoft CEO Satya Nadella has explicitly cited the book *1873* as essential reading, suggesting the parallel is not lost on the industry.
The current compute race exhibits the same escalating financial dynamics: companies are raising debt at deteriorating terms (86% of bonds issued in 2026 already trade above issuance yields, and bond cover has collapsed from 5x in February to below 2x today) to fund an implicit assumption that whoever controls the most compute will dominate AI and generate the cash to justify the spending. Google's recent leadership overhaul—removing DeepMind CEO Demis Hassabis and Chief Scientist Jeff Dean—is read by analysts as abandonment of frontier model development in favor of monetizing infrastructure through Google Cloud, where margins and growth are already accelerating (82% year-over-year growth, 36% margins). This suggests the real wealth may lie not in building frontier AI but in renting compute to others, a safer but less visible bet than the public posturing around model leadership.
Microsoft's ability to fund expansion through operating cash flow ($19.6 billion last quarter) rather than debt positions it distinctly from its peers, but the article implies this advantage is precarious: if compute capacity becomes truly scarce, the company with the most cash to deploy will dominate the market, compounding advantages and raising the risk that aggregate investment exceeds aggregate returns, reproducing the conditions of 1873.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Nvidia has partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise $500bn in…
Nvidia has signed letters of intent with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to…
Anthropic has signed the EU AI Act Code of Practice and will embed invisible watermarks in Claude-generated te…
Anthropic has agreed to pay $9.1 billion over 20 years to Riot Platforms Inc., a Bitcoin miner turned data cen…
Samsung is accelerating efforts to qualify its Taylor, Texas fabrication plant for 2-nanometer chip production…
Nvidia has partnered with six major investment and financial groups to mobilize more than US$500 billion in th…
The AI news that matters, in one minute each morning.
Sign up free