Google’s AI Return on Investment – Bristlemoon Research

Google has spent $132 billion on AI infrastructure over the past year, but Bristlemoon Capital estimates only $82 billion of that spending has actually gone into service.

Strip out the idle construction-in-progress and the math changes: a headline 16% post-tax incremental ROIC becomes an estimated 49%, according to the firm's recalculation using Google's own filings. Bristlemoon Capital, which may invest in securities featured in its newsletter, argues that as the remaining $123 billion backlog comes online through Q2 2027, Google's operating income could grow by more than $60 billion over the next twelve months, well above Bloomberg's consensus estimate of $43 billion.

Ticker: GOOGL (Alphabet)
Research Firm: Bristlemoon Capital
Report URL: https://www.bristlemoonresearch.com/p/googles-ai-return-on-investment?ref=shortreport.fyi
Position Disclosure: Bristlemoon Capital Pty Ltd states it "may invest in securities featured in the newsletter."

Thesis

Bristlemoon Capital argues that conventional ROIC math is masking how productive Google's AI spending already is, because it counts capex that hasn't yet gone into service.

• In-Service Accounting Gap: Of $132 billion in LTM capex, Bristlemoon estimates only $82 billion ($57 billion net of depreciation, $43 billion using average rather than period-end PP&E) was actually placed into service, turning a "naïve" 16% post-tax incremental ROIC into an estimated 49%.
• Idle Deployment Backlog: Google carries $123 billion of assets not yet in service, roughly four quarters' worth based on the past three years' pattern; about $110 billion net of depreciation is expected to activate by Q2 2027.
• Consensus Return Compression: Applying Google's LTM 60% pre-tax incremental ROIC to that backlog implies over $60 billion of NTM operating-income growth, versus Bloomberg consensus of $43 billion, a gap Bristlemoon says implies consensus is modeling ROIC falling into the low-40% range.
• Returns Improving, Not Fading: Incremental ROIC troughed in Q4 2025 and expanded over the following two quarters even as nominal quarterly capex kept rising.
• Uncounted Revenue Streams: The firm's above-consensus estimate excludes TPU hardware sales the CFO has said will be "material" in 2027, and earnings from a $920 million-per-month compute-rental agreement with SpaceX.
• Memory Inflation Drag: Even if $10 billion of the not-yet-in-service backlog is effectively wasted on rising memory costs, Bristlemoon estimates incremental returns would fall by only mid-single digits.
• Token Demand Ceiling: The report frames the real risk not as whether current compute demand exceeds supply, but as where AI token demand ultimately plateaus, calling that "the key piece of the entire ROI puzzle."
• Pricing Cushion: Frontier model API pricing is estimated to run gross margins in the 90s, potentially high 90s; in Bristlemoon's illustrative example, a 50% price cut from a 98% margin only pushes it down to 96%.

Catalysts

• Next filing visibility: Reporting on the estimated $33 billion of assets Google placed into service in Q2 2026, which Bristlemoon says should soon reveal the returns those specific assets generate.
• Backlog activation window: Continued placement of the $123 billion not-yet-in-service backlog, expected to largely complete by Q2 2027, is the key window for testing Bristlemoon's above-consensus operating-income estimate.
• TPU commercialization: TPU hardware sales becoming "material," per Google's CFO, are expected in 2027.
• SpaceX earnings disclosure: Disclosure of earnings tied to the $920 million-per-month SpaceX compute-rental agreement, which is not currently reflected in consensus estimates.
• Token economics shifts: Any shift in AI-token pricing, demand, or hardware/software efficiency could materially change future returns.

Company Response

The report does not indicate that Google was asked for comment, and no company response is included. The only external reactions cited are from commenters on Bristlemoon's own post and a reference to a prior CFO statement on TPU hardware sales; there is no direct quote or filing response from Google addressing the ROIC analysis itself.

Notable Details

• Bristlemoon estimates Google placed $33 billion of assets into service in a single quarter (Q2), calling it a live test of whether AI infrastructure delivers the returns management has promised.
• Incremental ROIC bottomed in Q4 2025 and then expanded for two straight quarters even as quarterly capex kept climbing, a pattern the report says cuts against the idea that AI spending is producing worsening economics.
• A commenter identified as "Melvin" noted that Google's purchase commitments and other contractual obligations more than doubled in a single quarter, from $332.4 billion to $811.0 billion.
• A separate commenter, "The Catalyst Shift," ran a comparable analysis on "SpaceX Colossus 1" using neocloud on-demand pricing and arrived at an estimated ROIC of 25% to 45% depending on capex assumptions.

"The greater uncertainty, and clearly the key piece of the entire ROI puzzle, is where token demand caps out."

Bristlemoon Capital, on the central risk to its bullish ROIC thesis: whether AI compute demand holds up as new capacity comes online.

FAQs

What does Bristlemoon's report say about GOOGL's AI spending returns?

Bristlemoon recalculates Alphabet's (GOOGL) incremental return on invested capital by excluding capex that has been purchased but not yet placed into service. Using that method, it estimates a 49% post-tax incremental ROIC for the trailing twelve months through June 2026, versus a 16% figure produced by a simpler calculation that counts all capex regardless of whether it's generating revenue.

Is Alphabet's AI infrastructure spending actually generating returns yet?

According to Bristlemoon's estimates, of the $132 billion Alphabet spent on capex in the trailing twelve months, only about $82 billion had been placed into service; the rest, roughly $123 billion including prior periods, sits on the balance sheet as construction in progress. The firm estimates about $110 billion of that backlog, net of depreciation, should be activated by Q2 2027.

Does Bristlemoon Capital have a financial stake in Alphabet?

Bristlemoon Capital's disclosure states it "may invest in securities featured in the newsletter," which includes Alphabet in this report. The firm is an Australian Financial Services Licensee (AFSL 552045) and separately promotes its Bristlemoon Global Fund to wholesale investors.

What is Google's $123 billion "not yet in service" backlog?

It refers to AI-related assets, such as data center and chip infrastructure, that Google has purchased and recorded on its balance sheet but that are not yet operational or generating revenue. Bristlemoon says this backlog has averaged roughly four quarters over the past three years before assets go live.

How does Bristlemoon's operating-income estimate compare to Wall Street consensus?

Bristlemoon estimates Google's next-twelve-month operating income could grow by more than $60 billion if its recent 60% pre-tax incremental ROIC holds as the backlog activates, compared with Bloomberg consensus of a $43 billion increase. The gap implies consensus estimates assume incremental ROIC falling into the low-40% range.

What is the SpaceX deal mentioned in the Alphabet AI report?

The report cites a compute-rental agreement between Google and SpaceX worth $920 million per month, which it says is excluded from current consensus operating-income estimates. Bristlemoon argues Google would not have entered the arrangement unless its own compute capacity were extremely constrained.

What is the biggest risk to Google's AI return-on-investment thesis?

Bristlemoon frames the primary uncertainty as where demand for AI tokens ultimately levels off, rather than whether current compute demand exceeds available supply. The report notes token demand is likely price-elastic, especially for near-frontier intelligence, and states the firm does not know where that demand could plateau.

Could rising memory chip prices hurt Google's AI returns?

The report acknowledges that memory-price inflation raises capex costs and could reduce returns unless Google Cloud passes those costs on to customers. Even in a scenario where $10 billion of not-yet-in-service assets reflect wasted spending on memory inflation, Bristlemoon estimates the hit to incremental returns would be only mid-single digits.

Disclaimer: This summary is not primary research and does not constitute investment advice. It is a brief overview of a detailed equity research report authored by the firm, organization, or source referenced in this article or at https://www.bristlemoonresearch.com/p/googles-ai-return-on-investment, which contains extensive evidence, regulatory filings, and analysis; readers are encouraged to review the full report there for a comprehensive understanding. The content provided in this publication is not authored or originated by us — we act solely as a distributor and do not endorse, verify, or take responsibility for the accuracy, completeness, or reliability of the information presented. This publication is for informational purposes only and should not be construed as legal, business, investment, or tax advice. Always conduct independent due diligence and consult qualified professionals before making any decisions based on the information contained herein. We disclaim all liability for any loss or damage arising from reliance on third-party content, and the views expressed are solely those of the respective source and do not necessarily reflect our own.