Oct 29, 2026

Google Unveils Gemini 4 Carbon as AI Spending Tests Big Tech

Google's new Gemini 4 Carbon model, built for long engineering and cybersecurity work, is going to vetted cyber defenders first. Meanwhile, Microsoft and Alphabet earnings showed booming AI cloud sales and the huge bills that come with them.

Listen to this update

Google released a new frontier model on Wednesday aimed at one of the most sensitive jobs in AI: finding and fixing security holes in software. Gemini 4 Carbon arrived on the same day Microsoft and Alphabet reported quarterly results showing how quickly AI cloud revenue is growing, and how much the companies are spending to keep up. Apple also launched its first touchscreen Mac.

Gemini 4 Carbon: a cyber model released to defenders first

Google describes Carbon as its strongest model yet. It is designed for long, multi-step engineering work, and Google says internal teams ran it for days at a time to trace and patch bugs across large codebases. It comes two weeks after Gemini 4 Argon, which Google opened to developers and subscribers earlier this month.

Most of the headline numbers are about security. Google reports that Carbon scored 88.2% on SEC-bench Pro, which measures finding vulnerabilities in real-world software, 71% on CWE-bench, which measures writing patches for known classes of weakness, and 54.8% on ExploitGym, which tests whether a flaw can actually be exploited. The company also says Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59. According to Google, Chrome's security team found it produced high-quality patches more than ten times as often as another lab's large frontier model, which Google did not name.

Because a model that can find exploitable flaws is useful to attackers too, Google is not releasing Carbon to everyone. For now it is limited to a selected group of cybersecurity defenders, and more trusted partners can apply through what Google calls the Fairwind Program. The company says it is also taking part in the U.S. government's voluntary pre-release testing of frontier models. Google plans to open Carbon to developers, businesses and consumers but has not given a date.

Google also highlights safety gains over Argon from its own testing. It reports zero successful attacks on the Gray Swan indirect prompt-injection benchmark and a drop of nearly 80% in high-severity misalignment alerts during internal coding work. Interesting Engineering notes that the multi-day autonomy and safety figures come from Google and have not been independently verified.

Introductory API pricing matches Argon: $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input. Google says prices will double after the introductory period.

Microsoft: Azure revealed at $33.2 billion

For the first time, Microsoft disclosed Azure's quarterly revenue in dollars: $33.2 billion for July through September, up 48% from a year earlier. That beat analyst expectations and the 44% to 45% growth the company had forecast in July, according to TechCrunch. Total revenue was $92.3 billion, up 19%.

Microsoft also reported about $10 billion in revenue from its commercial arrangements with OpenAI during the quarter, close to 11% of the total. The figure was $24.1 billion for all of fiscal 2026. OpenAI owed Microsoft $7.2 billion at the end of September, and Microsoft's stake in the ChatGPT maker slipped to about 24%.

The spending is heavy. The Motley Fool reports that Microsoft's capital spending reached $45.2 billion including finance leases, and free cash flow fell 27% from a year earlier. Even so, shares rose after hours.

Alphabet: cloud doubles, but cash runs negative

Alphabet's revenue climbed 25% to $127.6 billion, slightly below estimates. Google Cloud was the standout, with revenue up 103% to $30.7 billion and $11.1 billion in operating income. Search was weaker: growth slowed to 12%, down from 19% in the first quarter.

To make up for it, Google is putting more ads into its AI search experiences. It is bringing ads in AI Mode to international markets for the first time and introducing multi-turn ad formats in the U.S. CEO Sundar Pichai said Search queries grew at their fastest rate in five years, and that AI Mode queries are on average three times longer than traditional searches.

Investors focused on cash. Alphabet spent $51.4 billion on capital projects in the quarter, more than double the year before, and kept its 2026 forecast of $195 billion to $205 billion. It said 2027 spending will "increase significantly." Free cash flow came in at negative $9.1 billion, and long-term debt has risen to about $125 billion. TechCrunch reported that shares fell 8% after the results.

Apple ships its first touchscreen Mac

Apple has released a MacBook Pro with a touch-enabled OLED display, the first Mac with either feature. It comes in 14-inch and 16-inch sizes with M5 Pro or M5 Max chips, a Dynamic Island in place of the notch, up to 128GB of memory and up to 16TB of storage. A redesigned hinge keeps the screen steady when it is touched. Prices start at $2,899 and $3,399, about $400 more than comparable mini-LED models, according to MacRumors. Pre-orders are open, and the laptops launch November 6 alongside an OLED iPad mini and M6 updates to the base MacBook Pro and iMac.

Why it matters

Wednesday showed the AI industry from two angles. Labs are now building models capable enough that their makers are rationing access. The companies paying for that progress are earning record cloud revenue while borrowing and spending at a pace that is starting to worry investors. Amazon's results, due Thursday, will show whether AWS faces the same trade-off.