Gemini 4 Argon Doubles Its Price on December 1, and That Is the First Real Test of Whether Alphabet's $132 Billion Build Pays Back

Google Cloud backlog

$560.8B

Sep 30, 2026

Owner cash

-$10.7B

Twelve months to Sep 30, 2026, our calculation

Argon standard price per 1M tokens

$4 in / $20 out

From Dec 1, 2026

Call: Hold off, conviction 3 of 5.

Verdict: Gemini 4 does not improve the thesis yet. It does not disprove it either. The model is real, priced, and sold through Google Cloud, but it reached general availability on Oct 13, after the Sep 30 quarter closed, so none of its revenue is in the filings.

Owner cash is still negative, and Alphabet priced another six billion euros of debt yesterday. The thesis from our Oct 31 piece stands: Alphabet now funds its superintelligence build with outside money, and the payback date is not yet visible.

1. The question this piece answers

On Oct 31 we published "Alphabet Raised $138B From Outside Investors in Nine Months Because Its Own Cash No Longer Covers the Buildout." The core finding was simple.

In the first nine months of 2026, Alphabet spent $131.95B on property and equipment, more than its operating cash flow, and closed the gap with new debt, convertible preferred stock and new common shares ^[2].

Since then, Google has put a new frontier model family on sale. This piece looks at one product niche, Gemini 4 sold through Google Cloud, and asks a single question: does Gemini 4 make that capex pay back through Cloud revenue and backlog fast enough to turn owner cash positive?

If it does, the Oct 31 thesis softens. If it does not, Alphabet stays a company in a debt-funded build cycle.

2. What Gemini 4 actually is

Gemini 4 is a model family. Argon is the first member Google has released.

Gemini 4 Argon. Google DeepMind announced Argon on Sep 30, 2026, and first released it only to vetted cyber defenders through its Fairwind Program ^[5]. On Oct 13, 2026, Google made Argon generally available worldwide ^[6].

It is offered:

To developers through the paid tier of the Gemini API in AI Studio and Android Studio, under the model ID gemini-4-argon ^[6][8].

To enterprises through Gemini for business, the Gemini Enterprise Agent Platform on Google Cloud, and the Antigravity coding tool ^[6][7].

To consumers who pay for Google AI Pro or Ultra, in the Gemini app and AI Mode in Search ^[6].

Google Cloud's documentation lists Argon as generally available from Oct 13, 2026, with a retirement date no earlier than Oct 13, 2027. It supports a context window of 1,048,576 tokens and the same maximum output, batch inference, pay-as-you-go in standard and priority tiers, and Provisioned Throughput, which is reserved capacity bought ahead of time. Enterprise controls include data residency and customer-managed encryption keys, in global, US, EU, London and Singapore regions

3. How Argon compares with rival models

The best outside evidence so far comes from independent benchmark groups posting on X. It shows a model that is competitive at the top tier, not one that leads it.

Overall intelligence. Artificial Analysis rebuilt its Intelligence Index as version 5 on Oct 28, adding two new tests and moving most of the index onto private datasets ^[13]. On the earlier version of that index, users reading the Artificial Analysis leaderboard said Argon trailing Anthropic's Claude Opus 5.5 and roughly level with OpenAI's GPT-6 Astra ^[14].

We treat those placements as claims, not confirmed, because they come from users rather than from Artificial Analysis itself, and we have not found where Argon sits on version 5.

Knowledge work. Vals, the group behind the Vals Index that Google cites, posted that Claude Opus 5.5 leads its index with Argon close behind and within the margin of error, and that no model wins every part of it ^[15]. On a separate Vals legal benchmark, it said Argon scored highest on all three underlying tests but finished just behind Meta's Muse Wonder Full overall, again within the margin of error ^[16].

So Google's claim of first place on the Vals Index holds at launch but is a near tie, not a clear lead.

Vals Legal Index, top five models by accuracy and cost per test. Argon ranks second, 1.14 points behind Meta's Wonder Full, at a higher cost per test.

Image: Vals, via https://x.com/ValsAI/status/2113767926935265636

4. How Gemini 4 reaches the income statement

There are three routes from Gemini 4 to Alphabet's revenue, and only one of them is clearly measurable.

Route one: Google Cloud. API calls on the Gemini Enterprise Agent Platform, Provisioned Throughput reservations, and Gemini for business seats all book as Google Cloud revenue. This is the route that matters for the capex, because Cloud is where the compute is rented out.

In the third quarter, Google Cloud revenue was $30.71B, up 103% from a year earlier, with a segment operating margin of 36.21% ^[1]. That works out to about $11.1B of Cloud operating income in one quarter, our calculation.

Route two: subscriptions. Google AI Pro and Ultra subscribers get Argon in the Gemini app. That revenue sits in Google Services subscriptions, mixed in with YouTube Premium and Google One.

The filings do not split it out.

Route three: Search. Argon in AI Mode is meant to defend the advertising franchise. It earns nothing directly and adds serving cost.

So the clean test runs through Cloud. And Cloud's Q3 numbers predate Argon's general release.

Every dollar of Gemini 4 revenue so far will first show up in the fourth-quarter report.

5. The backlog is the best evidence, and it predates Gemini 4

At Sep 30, Alphabet disclosed $567.7B of remaining performance obligations, of which $560.8B belongs to Google Cloud. It expects to recognize a little over 55% within 24 months ^[2].

At Jun 30, Cloud backlog was $513.9B ^[3]. Net new Cloud commitments in the quarter were therefore about $46.9B, our calculation.

That backlog was signed while Argon was still unreleased. It reflects demand for compute, earlier Gemini models and TPU capacity, not Gemini 4 specifically.

What Gemini 4 can do is change two things in the fourth quarter: how much of the existing backlog customers draw down, and how much new commitment they sign.

A rough payback frame, our calculation and not company guidance: if 55% of $560.8B is recognized over 24 months, that is about $308B of revenue, or about $154B a year. At the third quarter's 36.21% Cloud margin, that would be roughly $56B a year of Cloud operating income.

Twelve-month capex was $159.80B. Even if every backlog dollar converts on schedule at today's margin, Cloud operating income covers about a third of one year's capex.

The backlog makes the build credible. It does not, on its own, make it pay back on a short clock.

6. Owner cash is still below zero

Owner cash is operating cash flow less capex, less capitalized software, less stock pay. Alphabet does not disclose capitalized software separately; it is inside capex.

For the twelve months to Sep 30, our calculation from the filings ^[2][4]:

On a nine-month basis, the Oct 31 piece put owner cash at negative $28.14B, against positive $30.83B in the same nine months of 2025 ^[2]. Buybacks were zero in 2026, while dividends continued.

For Gemini 4 to turn owner cash positive on its own, it would need to add more than $10.66B a year of extra cash on top of what the business already makes, while capex stays flat. Alphabet has given no sign that capex is flattening.

The 10-Q says proceeds from its stock and preferred offerings will fund capital spending to scale its superintelligence infrastructure and global compute (paraphrased) ^[2].

7. The Nov 2 offering says the build is not done

On Nov 2, 2026, Alphabet filed a preliminary prospectus supplement and a free writing prospectus for a new euro bond deal ^[11][12]. The pricing term sheet shows six billion euros of senior unsecured notes in seven series: a floating rate note due 2028 and fixed rate notes due 2028, 2030, 2032, 2035, 2039 and 2045, with coupons from 3.7% to 5.25%.

Settlement is Nov 5, 2026. Moody's rates the notes Aa2 and S&P rates them AA+, both stable ^[12]. Net proceeds before expenses sum to about 5.98 billion euros across the seven tranches, our addition from the term sheet ^[12].

The prospectus supplement says the money will go to general corporate purposes, which may include repaying existing debt ^[11]. It does not tie this deal to capex.

But it lands three weeks after Argon's general release and adds to the $85.22B of debt Alphabet issued in the first nine months, net of costs, alongside $42.87B of common stock and $19.06B of mandatory convertible preferred ^[2].

A company whose new product was already paying for its build would not need to keep borrowing at this pace. The credit ratings say lenders are comfortable.

The 2045 maturity says Alphabet is matching long-lived data centers with long-dated money, which is sensible, but it is also the funding pattern of a utility in a build cycle, not a software company throwing off cash.

8. The price change is the test to watch

Argon's price doubles on Dec 1, from $2 to $4 per million input tokens and from $10 to $20 per million output tokens ^[6][8]. That is the cleanest Gemini 4 signal investors will get.

If usage holds after the increase, Google has pricing power on its frontier model and every TPU it runs Argon on earns twice as much per token. That would be the first evidence that the capex earns software-like returns rather than rental returns.

If usage moves to cheaper models or rivals, Gemini 4 becomes a cost of staying in the race rather than a source of payback.

One other signal matters: Provisioned Throughput. Reserved capacity converts into contracted revenue and backlog, so a rising Cloud backlog in the fourth quarter after Argon's release would show Gemini 4 pulling in commitments.

9. What would change the call

Toward Buy:

Fourth-quarter Cloud revenue growth holds near the third quarter's 103% with margin at or above 36%, after Argon's full quarter on sale.

Cloud backlog grows by more than the $46.9B added in the third quarter.

Twelve-month owner cash turns positive in 2026's fourth quarter or 2027's first.

Alphabet discloses Gemini model or API revenue on its own.

Toward Avoid:

Argon volume falls after the Dec 1 price increase, or Google extends the introductory price.

Cloud margin falls as more revenue comes from selling TPU capacity at lower margin.

Debt and equity raises continue into 2027 with no owner cash recovery.

10. Valuation

We do not set a base, bear or bull value in this piece. Owner cash is negative at negative $10.66B over twelve months, so there is no positive anchor to value.

We will set one when owner cash turns positive.

11. Why Hold off, conviction 3 of 5

Gemini 4 is a credible frontier product, priced for enterprises, sold through every Google Cloud channel, and trained on Google's own chips. That vertical stack is Alphabet's real edge in the superintelligence race.

But the filings show no Gemini 4 revenue yet, the backlog that justifies the build was signed before Gemini 4 existed, owner cash is still below zero, and Alphabet borrowed again yesterday. Gemini 4 keeps the thesis alive.

It has not yet paid for it.

12. Predictions

1. Alphabet will not disclose Gemini 4 revenue on its own in the fourth-quarter 2026 report.

2. Google Cloud backlog at Dec 31, 2026 will be higher than the $560.8B at Sep 30.

3. Argon's standard price of $4 and $20 per million tokens will take effect on Dec 1 as announced.

4. Twelve-month owner cash will still be negative at Dec 31, 2026.

5. Argon will not hold first place on the Artificial Analysis Intelligence Index at any point before Dec 31, 2026.