Earnings Highlights: Alphabet (Google) | FY2026 Q2 — CapEx Climbs to $205B; the Next Battleground for Optical Interconnect Is "Across Data Centers"
This article is for technology and industry trend analysis only and does not constitute investment advice.
Key Takeaways
Demand is real: Google Cloud revenue rose 82% to US$24.8 billion, operating margin jumped from 20.7% a year ago to 35.6%, backlog hit US$514 billion, and the model API now processes 22 billion tokens per minute (up from 16 billion just a quarter ago). The market's biggest question of the past six months — "is AI capex actually turning into money?" — was largely put to rest this quarter.
More money, and where it flows: full-year capex guidance jumped from US$180–190 billion to US$195–205 billion, to "accelerate delivery of capacity." About 40% of it goes to data centers and networking equipment — and that 40% is the pool for optical modules, optical components and data center interconnect (DCI).
The most important line for STT: management unveiled the Virgo Network — linking millions of AI accelerators across multiple data centers into a "single supercomputer." This isn't cloud marketing; it's a structural event that pushes optical interconnect from inside the rack (scale-up) to between campuses and data centers (scale-across).
1. Three Signals to Remember First
First, this quarter's EPS is inflated — ignore it. Alphabet's net income and EPS surged mainly because "other income and expense (OI&E)" recognized US$98 billion, almost entirely unrealized gains on its equity portfolio — paper numbers, not earned from the core business. What to watch is operating income of US$40.8 billion (+30%), a 34% operating margin, and the Cloud line. Any reading that leads with EPS has the wrong focus.
Second, Cloud is the quarter's only star. Revenue +82%, operating income US$8.8 billion (tripled), operating margin 35.6%. More importantly, this is the first time it has begun recognizing revenue from "selling complete TPU systems to customers for installation in their own data centers." Google has formally moved from "renting compute only" to "selling hardware systems" — and that means far more for the supply chain than any single line in the financials.
Third, capex spending pushed free cash flow negative. Quarterly capex of US$44.9 billion drove free cash flow to −US$5.9 billion. A company sitting on US$242.5 billion in cash would rather let FCF go negative than slow its capacity buildout — that's a signal of conviction, not testing the waters.
2. Financials: Skip EPS, Look at Cloud and CapEx
Bottom line: The core ad business is steady (Search +17%, driven by retail and financial services), but this quarter's story is Cloud's acceleration and the capex increase. Network advertising at −1% is the only weak spot, but it's a small share and doesn't change the big picture.
3. Technology and Business Highlights: TPUs Start "Shipping Out," and Virgo Turns Data Centers into One Machine
For the optical communications supply chain, this call carried two signals far more substantive than the financial figures.
Signal 1: TPU systems "shipping out" into customer data centers. TPUs used to live only in Google's own facilities; this quarter, for the first time, Google recognized revenue from "delivering complete TPU systems into customer data centers" (the Blackstone deal being the example). The accelerator lineup is out in the open too: Google's in-house TPU 8t (training) / 8i (inference), plus NVIDIA's latest Vera Rubin platform. The implication: high-density, optics-heavy architectures like TPU pods are starting to spread to many more sites beyond Google. Optical interconnect demand is no longer tied to Google's own DCs; it spills out alongside the TPUs.
Signal 2: The Virgo Network — this is where optics plays at home. Sundar said plainly that Virgo "lets customers connect millions of AI accelerators across multiple data center sites into a single unified supercomputer." In optical terms: this is the last mile of scale-up (in-rack) → scale-out (in-cluster) → scale-across (between data centers), and physically that mile can only be covered by light — campus fiber, coherent long-haul interconnect, DCI. Once "one training job spanning several DCs" becomes a product, optical interconnect goes from supporting role to the foundation that makes it possible.
For a sense of scale: the model API processes 22 billion tokens per minute, up from 16 billion a quarter ago — throughput up 37% in a single quarter. That's the underlying reason capex can be spent and still not fill demand.
4. Management Outlook: $205B Is Just 2026; 2027 Will "Significantly Increase"
CFO Anat Ashkenazi raised full-year capex guidance from US$180–190 billion to US$195–205 billion, with the stated reason being "to accelerate delivery of capacity to meet growing demand." And 2027 capex will "significantly increase", with details to come later.
Reading the tone of the wording:
Management has said "supply-constrained" for several quarters running. Anat said outright that "demand still outpaces that investment." That isn't politeness; it's the industry reality that "money isn't the problem, capacity is" — mapping directly to tightness in optical components and modules.
In Q3, Google will "expand its use of third-party capacity" as a bridge, at the cost of near-term margin pressure, plus integration headwinds from the Wiz acquisition. So next quarter's Cloud operating margin may not repeat this quarter's 35.6% high — management has flagged this itself, so don't be caught off guard.
Only a small portion of TPU system sales revenue will be recognized this year; 2027 is when it ramps. In other words, 2026 is the year to "build capacity and inventory," with the real TPU hardware revenue contribution coming next year — which is why capex leads and revenue follows.
Comments