US Tech Earnings Diverge Sharply: Google Cloud Soars as Gemini 4 Carbon Debuts

Tech Giants' Earnings Season: Growth and Hidden Concerns Coexist

This week, US tech giants released earnings in quick succession, showing a clear divergence. Alphabet, Microsoft, and Samsung Electronics delivered sharply different results, reflecting both the widespread pains of the AI infrastructure investment phase and structural opportunities.

According to a roundup from the Bilibili tech news channel, Alphabet's third-quarter revenue reached $127.577 billion, up 25% year over year but below market expectations; net profit was $26.170 billion, down 34% year over year, and the stock fell more than 10% after hours at one point. The biggest highlight in the report came from Google Cloud—revenue surged 103% year over year to $30.709 billion.

However, the cost of aggressive expansion is equally striking: long-term debt soared 479% year over year to $125.1 billion, free cash flow was negative $9.07 billion for the quarter, and the investment portfolio posted an $8.66 billion loss. The company expects capital expenditure to exceed $300 billion in 2027. These figures show that growth in cloud and AI businesses is coming at the expense of short-term financial health.

Microsoft's Steadiness vs. Google's Aggressiveness

By contrast, Microsoft's first fiscal quarter performance was more balanced. Azure cloud revenue grew 48% year over year to $33.27 billion, beating market expectations; capital expenditure for the quarter was $45.2 billion, below the previously forecast upper limit of $50 billion; free cash flow reached $18.6 billion, far better than the market expectation of $8.54 billion. According to the source, among hyperscale cloud providers, Microsoft is currently the only company expected to maintain positive cash flow next year. After the earnings release, Microsoft's stock rose nearly 6% after hours at one point.

The divergence between the two cloud giants is worth noting: Google Cloud is growing faster, but its financial leverage and cash flow pressures are significant; Microsoft's growth is relatively moderate, yet it maintains an advantage in capital discipline. For readers focused on the investment logic of AI infrastructure, this difference means that when evaluating cloud providers, revenue growth alone is not enough—cash flow quality is equally a key indicator.

Free cash flow is the remainder of cash generated from business operations after subtracting capital expenditures, reflecting the cash a company can actually freely dispose of beyond what is needed to maintain and expand operations. For asset-heavy industries such as cloud and AI infrastructure, where capital expenditure is enormous, whether free cash flow is positive often says more about the health of the business model than revenue growth. Google Cloud's negative free cash flow of $9.07 billion for the quarter means its operating cash inflow for the quarter was insufficient to cover capital spending on data centers and chips, requiring reliance on external financing or existing cash—this is the direct backdrop for its 479% year-over-year surge in long-term debt.

Samsung Doubles Down on HBM4, Raises Capex by 60%

Samsung Electronics' semiconductor division operating profit surged to 106.5 trillion won in the third quarter, but slightly below the market's previously high expectations, and the stock fell 5% at one point in early trading. Facing fierce competition with SK Hynix in the HBM4 advanced memory segment, Samsung announced it will raise overall capital expenditure this year by about 60% to 84.4 trillion won (about $63 billion) to accelerate the expansion of advanced chip capacity.

This move shows that the memory chip race has shifted from capacity expansion to technological positioning in advanced packaging and high-bandwidth memory. Samsung chose to trade more aggressive capital spending for progress on HBM4, and short-term profit pressure is almost an inevitable cost.

HBM (High Bandwidth Memory) is a memory product that vertically stacks multiple layers of DRAM chips using through-silicon via (TSV) technology and packages them together with accelerators such as GPUs. Its bandwidth is far higher than traditional GDDR, making it a key component determining data throughput efficiency in AI training and inference clusters. HBM4 is the fourth-generation standard, further improving per-stack bandwidth and capacity. Because HBM manufacturing involves advanced packaging and yield control, capacity ramp-up is difficult; the competition between Samsung and SK Hynix is essentially a contest over advanced packaging processes and customer certification progress, which is why Samsung is willing to sharply raise capital expenditure.

Gemini 4 Carbon: Security Capabilities Become the New Selling Point

Alongside the earnings reports, Google officially launched its new-generation frontier model Gemini 4 Carbon, with a focus on strengthening complex programming, long-duration autonomous execution, and security attack and defense capabilities. Official data shows that the model achieves an 88.2% vulnerability detection score, doubles its ability to generate secure code, reduces the success rate of prompt injection attacks to 0%, and cuts high-severity security alerts by nearly 80%.

API pricing remains unchanged at $2 per million input tokens and $10 per million output tokens, and it is initially being opened to professional cybersecurity defense teams.

In terms of positioning, Gemini 4 Carbon is not an all-purpose model for general consumers, but is aimed at the vertical scenario of security attack and defense. Reducing the prompt injection success rate to 0% is a significant metric—such attacks have long been one of the main obstacles to deploying large models in enterprise environments. However, this figure comes from Google itself and still awaits independent third-party verification.

Prompt injection attacks refer to attackers embedding carefully crafted instructions into inputs to induce large models to deviate from their original system settings, leak sensitive information, or perform unintended operations. Such attacks have long topped the OWASP risk list for large-model applications and are a major concern for enterprises integrating large models into business systems. Reducing the injection attack success rate to 0% means the model has strengthened its ability to distinguish between instructions and data boundaries, but this metric depends heavily on the coverage of the test set; in actual deployment, it still needs to be evaluated together with peripheral defenses such as input filtering and permission isolation.

Undercurrents of M&A: Optical Communications and Open-Source Platforms

According to foreign media reports, chip designer Marvell Technology is in advanced talks to acquire data center optical communications equipment maker Lumentum, with a deal potentially announced as early as next week. Lumentum specializes in lasers and optical hardware used for high-speed data transmission in data centers, a key foundation for AI cluster interconnection. After the news, Lumentum rose 15% after hours at one point, while Marvell fell about 10% after hours.

Another piece of news has drawn more attention from developers: the US Department of Justice has opened an antitrust inquiry into Nvidia's planned acquisition of AI open-source platform Hugging Face for about $13 billion, as part of its broader review of the company's business practices. Regulators are focusing on the potential impact of the acquisition on the AI developer ecosystem and competition in the open-source market. Nvidia announced the deal in September and publicly promised to maintain the openness of the Hugging Face platform after the acquisition.

These two deals point to the same trend: the main battlefield of AI competition is extending from single-chip performance to interconnect hardware and the developer ecosystem. Optical communications determine cluster efficiency, open-source platforms determine ecosystem entry points, and regulatory intervention may reshape whether deals can go through.

Hugging Face is the world's largest open-source AI model and dataset hosting community, where developers can publish, download, and share model weights; many products from enterprises and research institutions are built on its ecosystem. If acquired by Nvidia, it would mean that the company controlling AI computing hardware also holds a key entry point to the developer ecosystem, potentially giving it influence over model distribution channels. For this reason, regulators are focusing not on the deal amount, but on whether the platform's neutrality can be maintained after the acquisition and whether models from other chipmakers can receive equal treatment.

Conclusion

From earnings data to model releases to M&A regulation, these threads together sketch the real state of the AI industry today: unprecedented scale of investment, lengthening return cycles, and a shift in competitive focus toward infrastructure and security capabilities. For investors, cash flow says more than growth; for developers, the ownership of the open-source ecosystem and the usability of security models will directly affect technology choices in the coming years.

One-line Summary

Tech giants' earnings diverge, with cash flow and ecosystem competition in the AI investment phase in focus.

Paragraph Summary

This week, earnings from Alphabet, Microsoft, and Samsung showed clear divergence: Google Cloud revenue surged 103% year over year, but net profit fell 34%, long-term debt soared 479%, and free cash flow turned negative; Microsoft Azure grew 48% and free cash flow far exceeded expectations, making it the only hyperscale cloud provider expected to maintain positive cash flow. Samsung's operating profit surged but slightly missed expectations, and it announced a roughly 60% capex increase to accelerate the HBM4 race. In the same period, Google released Gemini 4 Carbon, focused on security attack and defense; Marvell Technology is in talks to acquire Lumentum; and Nvidia's $13 billion acquisition of Hugging Face faces a DOJ antitrust inquiry. These threads together point to a trend of unprecedented AI industry investment, lengthening return cycles, and competition shifting toward infrastructure and security capabilities.