Penny Wise but Pound Foolish?
By Oliver Campbell │ ICM HPQC Fund │ November 2026
“Written by humans, please don’t blame the robots for our typos”
Like everyone else, we recently tore through Anthropic’s IPO document, the airport thriller of the season. There is plenty to like, but plenty of questions remain too. Amid all the civilisational commentary, one number stood out: US$895bn1 of capital commitments. Depending on your perspective, that is either unbearably big or acceptably small. Either way, someone has to pay for it. Internal cashflows can't, for now, public markets may lack the patience, and private markets may lack the depth. At HPQC we stay well away from AI software, but we are obsessed with the new technologies that will make AI commercially viable.
What is US$895bn?
There are two ways to read this number. On one reading, it is astronomically large, on a par with the TARP bailout during the global financial crisis or a year of the US defence budget. On the other, it looks like ‘only’ ~2% of US federal debt, ~20% of Nvidia’s market cap (correct at the time of writing), or one year of US government interest payments. Which side of the fence you land on will probably decide whether you think Anthropic’s future is rosy or racy.
How much is US$895bn?
Why are capital commitments so important?
Compute is the core input to Anthropic’s service, and its ability to stay at the frontier depends on securing it at the scale needed to train the next generation of models. The IPO document points to “Scaling Laws and the Predictability of Progress”, the observed link between more compute and better models, while admitting the relationship could plateau. Compute costs as a share of sales have risen 10 percentage points to 58% over the past two quarters, which has pushed the company back into a P&L loss. That is not a small cost.
Compute capacity is scarce and contested. Anthropic competes with some of the world’s largest technology companies, including its own hyperscaler suppliers. They can outbid it, get priority allocations from chip makers and cloud providers, and make long-term capital commitments that Anthropic can’t match.
To us at HPQC it seems clear that demand for compute continues to exceed available capacity (and our ability to add capacity with current technologies), and that Anthropic has previously asked cloud providers for more than they were willing to provide. Long-term contracts are how it reserves capacity before someone else takes it. Even signed contracts don’t guarantee delivery, though. The IPO document notes that providers can fall short on timing or volume because of their own capacity constraints and internal demand, so Anthropic has to line up supply from several partners rather than rely on one. Imagine Claude suffering a half-day outage once it is built into the economy. That could be fatal to its competitive position.
Long-term deals are also the price of getting capacity at all. Data centres and custom chips take years to build, so suppliers and landlords want multi-year take-or-pay commitments before they start. That is why the obligations stretch out to 2045 and why leases are signed years before they begin. The chart below shows the stack and how revenue and capex flow through it.
Where Anthropic sits in the stack
Lack of patience versus lack of depth
US$895bn over 20 years is ~US$50bn a year of what would effectively be operating costs. Annualised revenue could currently be as high as US$68bn, which suggests these costs may one day be digestible, especially if current growth holds. But if it doesn’t, because of competition or rising compute costs, Anthropic will have to raise financing, and that raises the question of where the money comes from.
Public equity markets are much deeper than private ones. Estimates put global public markets at roughly US$150trn, against US$7–20trn for private equity. There are about 50k listed companies and far more private ones.
The challenge with taking a five-year-old company public is that public markets are hyperactive and driven by news flow in the short term. Companies must also report quarterly, which is awkward for one that talks about decades ahead while still wrestling with profitability. Companies are right to think in decades, but modern stock markets often care more about the next quarter. Why should a public shareholder who might hold for a year, a month or even less tolerate near-term losses, even if Anthropic has a bright cashflow future somewhere out there?
On Nvidia’s earnings call last quarter, there was a glaring contradiction between CEO Jensen Huang and CFO Colette Kress on frontier AI labs and how they are financed.
Nvidia CEO Jensen Huang said this3
Nvidia CFO Colette Kress said this4
These two statements can’t both be true at the same time. Either ROIC is above 100%, as the first quote implies, or Colette is right and the labs are growing faster than their cashflow and balance sheets can support. Investors in a frontier AI lab like Anthropic will need to decide what payback window is likely and tolerable. In a world of constrained compute and fierce AI competition, we find that very hard to judge.
Why we prefer the stack to frontier AI
For us at HPQC it comes down to the picks-and-shovels argument. We can’t tell you which AI lab will survive or win the race, but we are confident that AI is here to stay and that demand for compute will keep growing against a backdrop constrained by physics.
Electric vehicles are a useful analogy. We can’t tell you which car maker will win the EV market share battle over the next few decades, whether Tesla, BMW, Volkswagen, Mercedes, Audi, BYD or a newcomer like Jaecoo. We can assume, though, that EVs will keep taking share from internal combustion engine (ICE) cars. That makes investing in the supply chain, whether chargers, inverters, battery technology or power ICs, feel more predictable than guessing which brand to back.
Similarly, whichever AI labs come out on top, we believe AI as an industry will keep growing. The key bottleneck to any AI is compute and the supply of watts. Those watts need to be directed as efficiently as possible to generate the most tokens, and those tokens need to be allocated as efficiently as possible to deliver the most powerful inference. So rather than guessing between ChatGPT, Claude, xAI, DeepSeek, and others, we would rather invest in the new companies that will help generate the most tokens per watt. These sit in emerging sectors including power, photonics, advanced memory, advanced packaging, cooling, sparse computing and others.
Which brings us back to who pays. Someone will fund the US$895bn, whether through public equity, private capital, debt or the suppliers themselves, and the terms will depend on how patient that money is. We can’t say who ends up footing the bill. We can say that whoever does will be buying power, photonics, memory and packaging, and we would rather own a piece of that than guess which lab it ends up serving. Rather than panicking about recent newsflow-driven public market gyrations feeding back into demand, we believe that the more likely long-term risk to our approach is overbuilding: if capacity runs ahead of demand, the stack eventually suffers too. Ongoing physical constraints make that unlikely for some time, but it is the thing we watch.
Oliver Campbell
Investment Principal of the ICM HPQC Fund
MAS Licensed Representative, ICM Global Funds Pte Ltd
November 2026
Sources:
[1] https://www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm pp.6 “Our approach to Compute”
[2] Ibid, pp.160 “Quarterly Consolidated Statements of Operations”
[3] Nvidia 10Q 2QFY27, Pitchbook
[4] Ibid
Important Note:
The information in this article should not be considered an offer or solicitation to deal in the ICM HPQC Fund (Registration number T22VC0112B SF003) or ICM HPQC Fund 2 (Registration number T22VC0112B SF007) (the “Sub-funds”). The information is provided on a general basis for informational purposes only and is not to be relied upon as investment, legal, tax, or other advice. It does not take into account the investment objectives, financial situation, or particular needs of any specific investor. Investors should seek relevant professional advice before making any investment decision. The information presented has been obtained from sources believed to be reliable, but no representation or warranty is given or may be implied that it is accurate or complete. The Investment Manager reserves the right to amend the information contained herein at any time, without notice. Investments in the Sub-funds are subject to investment risks, including the possible loss of the principal amount invested. All forms of investments carry risks, including the risk of losing all of the invested amount. Investors should read the prospectus before deciding whether to acquire the units in the Sub-funds. The value of investments and the income derived therefrom may fall or rise. Past performance is not indicative of future performance. This document is intended solely for institutional investors and accredited investors as defined under the Securities and Futures Act 2001 of Singapore. The whole or any part of this work may not be reproduced, copied or transmitted or any of its contents disclosed to third parties without ICM Global Fund’s express written consent. This advertisement or publication has not been reviewed by the Monetary Authority of Singapore.
ICM HPQC Fund and ICM HPQC Fund 2 are registered Sub-Funds of the ICMGF VCC (the VCC), a variable capital company incorporated in the Republic of Singapore. The assets and liabilities of ICM HPQC Fund and ICM HPQC Fund 2 are segregated from other Sub-Funds of the VCC, in accordance with Section 29 of the VCC Act.
Get to know us better:
Comments