Investor Note · 21 May 2026
OpenAI ARR went from $2B to $25B in 24 months. Anthropic from $300M to $30B. Hyperscaler margins are expanding through the capex cycle, not contracting.
Invesense Research · Company filings via Yahoo Finance · Sacra · The Information · SaaStr · 2018–2027E
$357B
2025 hyperscaler capex
1.7× the 2000 telecom peak
+$25B
OpenAI ARR, Feb 2026
up from $2B in 24 months
46%
Microsoft operating margin
expanding through the cycle
2.1x
Microsoft OCF to capex
funded from cash flow
Hyperscaler capex hit $357 billion in 2025 across Microsoft, Alphabet, Meta, and Amazon. The 2026 consensus is $500 billion. The 2027 number is $625 billion. Cumulative 2024–2027 sits north of $1.7 trillion — more than three times the entire 1996–2001 telecom buildout. The number is real, the number is large, and the number is being funded out of operating cash flow. The bill is already being paid.
01
Four companies are doing this. Microsoft spent $65B on capex in fiscal 2025. Alphabet spent $91B. Amazon spent $132B, including fulfillment but with cloud infrastructure stacked inside it. Meta spent $70B. The combined number tripled in three years and quintupled in five. The peak of the telecom buildout in 2000 reached $213B in inflation-adjusted dollars. Hyperscalers crossed that threshold in 2024. In 2025 they ran 1.7× faster than the loudest historical analogue.
The bear case begins here — the comparison to the telecom bubble, the warning that capacity always overshoots. The bear comparison misses what comes next.
02
Telecom fiber laid in 1998 did not generate meaningful revenue until streaming and cloud arrived a decade later. The AI cycle is not waiting. OpenAI ARR went from $2B at the end of 2023 to +$25B in February 2026. Anthropic moved from $300M to +$30B over the same window. Google Cloud revenue compounded from $33B in 2023 to $58B in 2025. AWS reached $115B. Eight of the Fortune 10 are Anthropic customers. Over 500 companies pay more than $1M per year. Salesforce took 15 years to reach $10B in ARR. OpenAI took 24 months.
On a log scale, the AI labs are not climbing a normal adoption curve — they are climbing a steeper one than any enterprise software cohort on record. The demand is not a projection. It has already arrived, and it is being paid for.
03
This is the part that ends the comparison. Microsoft, Alphabet, Meta, and Amazon each clear $200B+ in annual revenue. Walmart at $648B operates at 4% margin. Berkshire at $365B operates at 12%. IBM at the 2000 peak ran at 13%. Microsoft is at 46%. Meta at 41%. Alphabet at 32%. Cisco at the dot-com peak operated at 25% margins on $19B of revenue. Microsoft today runs higher margins on fifteen times the revenue. The capex cycle is not breaking the margins. The capex cycle is being absorbed by the margins.
| Company (FY25) | Revenue | Capex | OCF | OCF / Capex | Op Margin |
|---|---|---|---|---|---|
| Microsoft | $282B | $65B | $136B | 2.1x | 46% |
| Alphabet | $403B | $91B | $165B | 1.8x | 32% |
| Meta | $200B | $70B | $116B | 1.7x | 41% |
| Amazon | $717B | $132B | $140B | 1.1x | 11% |
| Nvidia (FY26) | $216B | $6B | $103B | 17x | 60% |
Every one of these companies funds its capex out of operating cash flow, not balance-sheet leverage. Microsoft generates $2.10 of operating cash for every $1 of capex; Nvidia, $17. The telecom operators of 2000 borrowed to build. The hyperscalers of 2026 are paying as they go.
Our View
The AI capex bill is already being paid. OpenAI ARR climbed from $2B to $25B in 24 months. Anthropic from $300M to $30B in 18. Google Cloud and AWS are expanding margins through the capex cycle, not compressing them. No company in modern history has operated at hyperscaler margins at hyperscaler scale.
This is the cleanest infrastructure cycle ever recorded. We are positioned long Mag 7 quality. The question is not whether the math works. It already does.
Speak with our investment team
For a deeper discussion on how these themes apply to your portfolio, we welcome a conversation.
Invesense Asset Management Ltd. is regulated by the Dubai Financial Services Authority (DFSA). This material is for informational purposes only and does not constitute investment advice or an offer to buy or sell any securities. Past performance is not indicative of future results. Financial data sourced from company filings via Yahoo Finance. Private AI lab revenue reported by Sacra, The Information, and SaaStr. Telecom buildout benchmark from MOI Global, the Richmond Federal Reserve, and historical industry data. Computations by Invesense Research.
← Back to Insights