The market consensus is that the high growth of optical modules depends entirely on increased capital expenditure by North American cloud providers. In the Q4 earnings reports of Microsoft, Meta, Amazon, and Google at the end of January 2027, which numbers would be the first to overturn this consensus?

The market consensus is that the high growth of optical modules depends entirely on increased capital expenditure by North American cloud providers. In the Q4 earnings reports of Microsoft, Meta, Amazon, and Google at the end of January 2027, which numbers would be the first to overturn this consensus?

What flips first is not total capital expenditure, but the "servers and network equipment" portion within it. The money for buying land, building facilities, and hooking up power does not reach the optical module makers' books at all.

The quarterly reports those four companies are due to release at the end of January 2027 are not out yet; the most recent period is the third quarter, released October 28 to 29 (Microsoft's is FY27Q1). None of the numbers below are theirs — but the question here is which number flips first, and that can be determined now.

Meta is the first to report at the end of January

Microsoft and Meta report at the end of January; Google and Amazon wait until early February. Of the four, only Meta writes a full-year absolute range for the next year into its press release: in 2026 it gave 1,150 to 1,350 yi dollars, raised three times within a year to 1,350 to 1,450 yi. The end-of-January release will give the 2027 range directly.

Google gives direction on the call, not numbers. The CFO's exact words in the third quarter were:

"We continue to expect our CapEx to increase significantly in 2027, and we'll provide further details on our Q4 call." (2026-10-29)

The absolute number has to wait for the Q4 call.

Amazon only gives 2026 "about 2,200 yi" and says "meaningful increases" for 2027, with no number. Microsoft gives about 1,700 yi for calendar 2026 (including finance lease principal), and for the next year only quarterly commentary.

So the word "first" has an owner: Meta's 2027 range is the first number among the four from which a year-on-year figure can be calculated.

The table uniformly takes "purchases of property and equipment" from the cash flow statement, excluding finance lease principal — Meta's 349.1 yi in the press release and Microsoft's 452 yi on the call both include leases, and comparing them directly would overstate.

Google prints the quarterly series directly in its earnings release; the slope looks like this:

Capex is accelerating — that is the entire body of evidence for the consensus. Two things need asking: whether this line can continue, and how much of it turns into optical modules.

A crack has already appeared

In the third quarter of 2026 the four companies totaled 2,019 yi dollars, +107.6% year-on-year. In the same quarter, on the same chain, the revenue growth of three optical module makers differed by 140 percentage points.

Quarterly figures are derived by subtracting the half-year report cumulative from the third-quarter report cumulative; gross margin is back-calculated the same way.

Eoptolink was -13.0% quarter-on-quarter; as revenue fell, inventory rose from 66.0 yi to 174.1 yi, and inventory days stretched from 187 days to 270 days. It is not that it has no orders, but that the orders did not arrive on schedule.

Zhongji Innolight's November 3 investor relations record explains the reason clearly:

"The vast majority of customers have not cut demand or orders; individual customers have product structural adjustments, and some customers' 1.6T demand has even increased compared with before."

The total is fine; the structure is moving. The same capex is distributed unevenly across customers and products, and landing at different module makers it becomes the difference between +171.8% and +80.3%.

Tianfu goes the other direction: gross margin rose from 53.7% to 65.1%, 11.4 points. Zhongji Innolight rose only 3.8 points, Eoptolink fell 0.7 points. The closer to assembly, the thinner; the money is moving upward.

2027 is already sold out

In the same record there is a more critical sentence: "The vast majority of the company's 2027 orders already cover the full year."

If this holds, the end-of-January capex guidance cannot hit 2027 revenue; the orders are already locked. What it actually prices is 2028 — Zhongji Innolight says that for 2028 "at least one key customer has already given clear demand guidance," and other customers are all waiting for this round of earnings at the end of January.

The transmission chain is therefore: end-of-January capex guidance → 2028 orders placed in February to March → contract liabilities move first in 2027Q1 → revenue moves next in 2027Q2. Contract liabilities lead by one quarter, revenue lags by one quarter, with the production cycle in between.

At the end of the third quarter, Zhongji Innolight's contract liabilities were 0.11 yi and Eoptolink's 3.05 yi — both absolute values are very small; these two do not rely on advance receipts to lock capacity, so watching contract liabilities is less useful than watching inventory and receivables.

Two numbers must be watched together

The first is Meta's 2027 range. 2026 is 1,350 to 1,450 yi, close to doubling year-on-year. If 2027 is given only around 1,600 yi, growth falls from doubling to the teens.

The second is the share of Google's servers and network equipment. At the February 2026 call it said about 60% of 2025 went to machines and 40% to long-cycle assets, with 2026 roughly the same; at the October call this became "about 65% is servers and network equipment, 35% is data centers."

Only if the share moves up can optical modules outrun capex; if this share reverses, capex can rise all it wants and it has nothing to do with optical modules.

Capex up 30%, optical modules up 60% — the consensus is falsified, but in a good direction; capex up 30%, optical modules up only 15% — that is what this narrative looks like when it dies.

Under what circumstances am I wrong: if the lower bound of the 2027 range given by Meta is above 1,700 yi USD, and the 2027 figure given by Google is above 2,400 yi, and Google explicitly states that the share of servers and network equipment is still 65% or higher, then "capex growth falls first, and optical modules fall along with it" does not hold.

There is also an earlier one: by the end of January, those companies' actual Q4 spending had already fallen into the lower half of guidance; there is no need to wait for 2027 guidance, the crack tears open on the spot.

Sources: Microsoft 8-K Exhibit 99.1, 2026-10-28, 0001193125-26-436122; FY27Q1 earnings call, 2026-10-29. Meta 8-K Exhibit 99.1, 2026-10-21, 0001628280-26-066700; 2026Q3 earnings call, 2026-10-21; 8-K Exhibit 99.1, 2026-01-28, 0001628280-26-003832. Amazon 8-K Exhibit 99.1, 2026-10-29, 0001018724-26-000038; 2026Q3 earnings call, 2026-10-30. Alphabet 8-K Exhibit 99.1, 2026-10-28, 0001652044-26-000080; 2026Q3 earnings call, 2026-10-29; 2025Q4 earnings call, 2026-02-04. Zhongji Innolight 2026 Third Quarter Report, 2026-10-30, Cninfo 1225643953; Investor Relations Activity Record Form 2026-010, 2026-11-03, Cninfo 1225656565. Eoptolink 2026 Third Quarter Report, 2026-10-30, Cninfo 1225642947. TFC Communication 2026 Third Quarter Report, 2026-10-29, Cninfo 1225634414.