Alphabet Cuts Buybacks to Pivot Toward Data Centers, Wall Street Fears CapEx Spiraling Out of Control
nashnova research
Alphabet has gone two straight quarters without a single share buyback — the first time since 2017 — while committing roughly $200 billion in annual capex to data centers; spending now exceeds operating cash flow, and Wall Street worries the AI buildout is devouring shareholder returns.
Where is the money going — buybacks stopped, data centers burning cash?
Alphabet conducted zero share buybacks for two consecutive quarters, the first such streak since 2017.
Meanwhile, the 2026 capex budget sits at roughly $200 billion, and CFO Anat Ashkenazi said on the last earnings call that next year's spending will "increase significantly."
This means → the company is channeling virtually all capital that could have gone back to shareholders into AI data centers — the buyback halt is the clearest signal.
How does Alphabet fill the gap — debt, equity, convertibles?
Capex already exceeds operating cash flow; Wall Street expects this shortfall to persist through 2028.
Over the past nine months Alphabet added $77 billion in new debt across multiple currency markets. In the first half of 2026 it raised $30 billion through stock issuance and another $19 billion via mandatory convertible preferred shares.
In plain terms = the company's own earnings can no longer cover the data-center buildout, so it is borrowing and selling equity to close the gap — with more stock sales planned in Q3.
Anthropic — customer, competitor, or the biggest wildcard?
Anthropic filed its IPO prospectus this week, listing Google simultaneously as a competitor, a major shareholder (over 5%), and one of its most important suppliers.
Anthropic has a cloud-services agreement with Google Cloud worth $114.5 billion running through 2033, plus a TPU (tensor processing unit — Google's custom AI chip) lease contract valued at $125–150 billion for 2028–2031.
This means → Google Cloud and TPU revenue are heavily tied to Anthropic's orders — if Anthropic shifts supplier strategy after going public, a key revenue pillar comes under pressure.
The SpaceX stake — $99 billion gain last quarter, a potential $11 billion hit this quarter?
Alphabet holds 551 million shares of SpaceX, valued at roughly $94 billion at the end of last quarter.
When SpaceX went public, Alphabet booked $99 billion in unrealized gains. But SpaceX stock fell 11.7% in Q3; if unsold, the position would produce roughly $11 billion in unrealized losses.
In plain terms = these gains and losses are all paper figures — nothing is real until shares are sold — yet they swing net income and EPS violently each quarter, making it hard for investors to see what the company actually earned.
Can anyone still compare EPS — fines, paper gains, and paper losses all in the mix?
In Q3 2025, Alphabet recorded $11 billion in unrealized equity gains, partly offset by a $3.5 billion EU fine, yielding EPS of $2.87.
The consensus estimate for the current quarter is $3.02, but year-over-year comparability has deteriorated sharply.
This reflects a deeper problem: when a company's income statement is simultaneously buffeted by massive paper swings and one-off fines, traditional EPS comparisons lose most of their meaning.
What is the market really watching — is the AI spending spree worth it?
Beyond capex, Alphabet spent $56 billion on acquisitions and private securities in the first half, most of it on Anthropic shares.
As of June 30, the balance sheet still holds ample cash and marketable securities — but the company also disclosed $900 billion in future supply and lease commitments.
This means → the core question is singular: can Alphabet prove that the relentless data-center buildout will eventually convert into profits, rather than simply becoming an ever-larger bill?
市场有风险，内容仅供研究参考，不构成投资建议。