Google and Meta Face the Dilemma of Using Computing Power Internally vs. Selling It
Alina Collins
Google and Meta face the same strategic bind — keep their compute capacity for internal AI training, or sell it to enterprise customers for cash flow relief. Google's operating cash flow turned negative for the first time in Q2; Meta's fell 91% year-over-year. Massive capex is now hitting the balance sheet, and how each company allocates its compute will shape both cloud revenue growth and AI competitiveness.
What did Zuckerberg actually say?
Meta currently has no business selling compute externally, but Zuckerberg said publicly for the first time that it is "in the plans."
His bluntest line: "Selling all the compute and taking the short-term profits would be foolish."
This means → Meta's priority order is clear — a "significant proportion" of compute stays in-house for AI model training and core products. External customers come second.
In plain terms = Zuckerberg sees compute as a compounding asset. Monetizing it too early is selling the future at a discount.
How bad is the cash flow pressure?
Google's Q2 operating cash flow turned negative for the first time. Meta's cash flow dropped 91% year-over-year — massive capex is directly hitting the financials.
Both Google and Meta raised full-year capex guidance slightly. Google hinted 2027 spending could climb further.
This reflects a paradox: spending keeps rising, yet every company believes pulling back now means falling behind. By contrast, Microsoft held its existing capex guidance unchanged — a noticeably different pace.
Why isn't there enough compute to go around?
Microsoft CFO Hood said on the earnings call that cloud business "customer demand continues to exceed available capacity."
Bernstein analyst Shmulik called Google's compute allocation a multi-objective optimization problem: protect its position at the AGI research frontier, defend the search moat, and retain the developer ecosystem — each one demands compute.
His exact words: "If you don't have enough compute for enterprise customers, they'll go straight back to Amazon or Microsoft."
In plain terms = the supply is fixed, internal teams and external clients both want it, and shorting either side carries a cost.
How is Google trying to solve this?
Google announced a two-track approach: buy more third-party compute to serve current customers while continuing to build out its own infrastructure.
On the chip side, Google is deploying its Tensor Processing Units (TPUs) — its custom-designed AI chips — into partner data centers to free up additional capacity.
Pichai set the priority on the earnings call: TPUs exist first to keep Google competitive at the AGI research frontier. "That is the foundation of everything we do."
What is the core tension in this game?
Keep too much compute internally → cloud customers leave, revenue growth stalls.
Sell too much externally → internal AI training is starved, long-term competitiveness erodes.
This means → there is no fixed "right ratio" in the short term. Google and Meta can only adjust dynamically — and where that balance point lands will directly shape both companies' cloud revenue and AI progress over the coming quarters.
Content is for reference only, not financial advice.