Alphabet Raises 2026 Capital Spending Forecast to $195–$205 Billion; Q2 Free Cash Flow Was Negative $5.85 Billion and Shares Fell About 6–7%

The Quarter Google Made Nothing

For twenty-two years, since the day it went public in 2004, Alphabet's core business has done the same reliable thing: bring in more cash than it spends. On July 22, 2026, that streak broke. Alphabet reported second-quarter revenue of $119.8 billion, up 24% from a year earlier, and Google Cloud revenue up 82%^[1][3]. Yet free cash flow — the cash left over after running and building the business — came in at negative $5.85 billion^[13][14]. Shares fell roughly 3% to 5% after hours that evening, then dropped another 6% to 7% the next trading day^[5][6].

Those two facts sit next to each other awkwardly on purpose. A company can grow fast and still burn cash, if it's spending on its own future faster than it earns money today. That's exactly what Alphabet says happened. The question nobody can fully answer yet is whether the spending is building something durable or just replacing itself as fast as it's built.

Where the Money Actually Went

The number driving the reaction is capital expenditure — money spent on physical things like servers, buildings and networking gear. Alphabet spent $44.9 billion on this in the second quarter alone^[4]. That's more than the $39.1 billion its operations generated in cash over the same three months^[13]. Subtract one from the other and you get the negative $5.85 billion free cash flow figure that rattled investors^[13][14].

Alongside the earnings report, Alphabet raised its full-year 2026 capital spending forecast to a range of $195 billion to $205 billion. That's up from the $180 billion to $190 billion range the company gave back in April^[2][5]. CFO Anat Ashkenazi delivered the new guidance on the earnings call, alongside CEO Sundar Pichai^[4][17]. Roughly 60% of that money is going to servers, and about 40% to data centers and networking equipment^[2][10].

Alphabet isn't funding all of this from its own cash pile anymore, either. In June 2026, before this earnings report, the company raised $49.6 billion through a mix of stock and mandatory convertible preferred shares, earmarked for AI infrastructure and compute^[9]. That's a meaningful shift: outside investors are now underwriting part of the AI buildout, not just Alphabet's own profits.

The Case for the Spending: A Backlog It Can't Fill

Alphabet's own explanation leads with a number it wants investors to sit with: a $514 billion cloud backlog. That's the total value of contracts customers have already signed with Google Cloud, for work the company hasn't yet delivered^[3]. In management's telling, that backlog is the whole story. Customers want more computing power than Google can currently supply, so the constraint isn't demand — it's capacity^[2][4].

Data centers take years to build and then earn revenue for years after that. Judging one on a single quarter's cash flow, in this framing, is like judging a house purchase on closing day. Management's competitive argument is asymmetric: losing enterprise cloud customers to Microsoft, Amazon or OpenAI-linked providers because Google couldn't supply enough compute would be a permanent loss. Building too much, by contrast, is a cost that fades^[2].

That's not a fringe read. Trade publications aimed at the data-center and infrastructure industry covered the raised guidance with almost no hint of alarm. DatacenterDynamics, a UK-based outlet whose advertisers are the very companies building this infrastructure, ran the number — $195 billion to $205 billion — without mentioning the cash-flow reversal or the stock drop at all^[10]. Gulf outlets like The National framed it as a race Alphabet is running, not a risk it's taking^[9].

The Case Against: A Bill That Hasn't Come Due Yet

The skeptics don't dispute that Google Cloud is growing or that the backlog is real. Their argument runs through an accounting concept called depreciation, and it's worth slowing down on because the whole dispute hinges on it. When a company buys equipment like a server, it doesn't record the full cost as an expense the day it's bought. Instead, it spreads that cost across the years the equipment is expected to stay useful, chipping away a little at reported profit each year.

That "useful life" estimate is a judgment call, not a physical fact, and it changes the numbers a lot. A $6 billion server farm written off over six years costs $1 billion a year in reported profit. Write it off over three years instead, and it costs $2 billion a year — same servers, same spending, a very different-looking bottom line. Alphabet currently depreciates its servers over roughly five to six years^[11].

Critics, including investor Michael Burry, argue that's too generous, because each new generation of AI chip is so much more efficient that the old ones become obsolete in two to three years, not five or six^[11][12]. If Burry is right, hyperscalers are understating their depreciation costs and flattering today's profits, with the real bill landing later. His own published estimate puts that understatement at roughly $176 billion across hyperscalers between 2026 and 2028^[11].

The skeptics also point to a real precedent: two major platforms reached opposite conclusions about similar hardware within weeks of each other in early 2025. Meta extended its server useful life from four years to five and a half. Amazon went the other way, shortening the useful life of a subset of servers and networking gear from six years to five^[11]. Same industry, same kind of equipment, opposite calls — which the skeptics say proves the assumption is a choice, not a measurement.

What the Reaction Reveals About Who's Watching

The stock's fall despite a strong quarter is itself a signal about what investors have stopped rewarding. CNBC covered the beat first, then the sinking stock, with a follow-up piece describing the report as "panned" by the market and warning that hyperscalers broadly now face tougher scrutiny on capital spending^[4][7]. The Motley Fool asked whether the drop was an "overreaction" investors should buy into — a framing that leans toward its own answer, consistent with an outlet whose business is picking stocks^[6].

The selloff wasn't contained to Alphabet. Meta and Snap fell the same day, and total 2026 AI capital spending across the four biggest U.S. tech platforms is now projected at roughly $600 billion to $725 billion^[19][20]. That scale is why suppliers of chips, power and networking equipment see this differently than Wall Street does: a $15 billion increase in Alphabet's guidance lands directly on their own revenue lines. Analysts have already named specific companies expected to benefit^[20].

There's a third audience that barely shows up in any of this financial coverage at all: the places where the data centers actually sit. Google's own reporting shows its data centers used 30.8 million megawatt-hours of electricity in 2024, roughly double the level from four years earlier^[8][9]. Consumer advocates in states hosting new sites argue that without changes to how utilities set rates, the cost of new power plants and transmission lines lands on ordinary electric bills while the benefit flows to the tech company and utility shareholders^[9]. Google's answer is that it's bringing new supply with it — one Minnesota project alone is tied to about 1,900 megawatts of new clean energy, including 1,400 MW of wind, 200 MW of solar and 300 MW of storage^[9]. Both things can be true at once: new clean power getting built, and local bills still climbing. That fight plays out in state utility commissions, on a much slower timeline than an earnings call.

What's Still Open

Three weeks out, some of the sting has faded. One market analysis on August 3 put Alphabet's stock up 14% from its post-earnings low, and shares closed at $357.52 on August 10^[15][16]. That recovery doesn't settle anything — it just means the market's initial reaction wasn't the market's final word.

The real test won't arrive on any earnings call. It's whether the chips and data centers Alphabet is buying now are still earning their keep in 2028, when this year's spending starts showing up as depreciation on the income statement starting in 2027^[18]. Until then, both readings of the same $205 billion stay equally defensible: a company meeting real demand it can't currently fill, or a company on a treadmill it can't yet prove is worth the pace.