Compute in Orbit, Safety in Doubt, and a Capital Frenzy: Global Tech Enters a Turbulent Season
Over the past two weeks, the global high-tech industry has experienced a series of significant events. Google sends AI chips to space, OpenAI fires researchers for mishandling sensitive information, Tencent invests $7 billion to lease Oracle’s cloud computing power, and Apple plans to release multiple major hardware products in two waves within a month. All of these events combine to show a more intense technology competition, an escalation of power, and a redrawing of industry rules.
Google’s Orbital Compute Bet
In early October, Google launched into space a prototype satellite carrying four Tensor Processing Units (TPUs). It went up aboard SpaceX’s Transporter-18 mission, formally marking the beginning of the testing stage of its Project Suncatcher.
The device was controlled and communicated with. In the coming weeks, engineers will focus on collecting data on how TPU chips perform under the conditions of space radiation, extreme temperature fluctuations, and physical shock, to optimize future designs.
The concept is compelling. The satellites, in low Earth orbit, can receive an almost continuous supply of sunlight, producing a total of eight times as much power as terrestrial systems. The goal is to establish several satellite networks that will eventually interconnect, allowing a large number of tasks to be performed in space.
If this vision of the future should come true, it would change the whole situation of data centers on Earth: the cost of land, energy, and cooling is prohibitive, whereas in space there is theoretically no limit to the energy and low temperatures of the environment. But of course, radiation damage to chips, the impossibility of on-orbit maintenance, and the communication latency across satellite constellations are obstacles that have to be surmounted, and so, for the moment, Google is looking more like a long-term technology option for the next decade than an immediate commercial solution.
They hope to launch two more experimental satellites in 2027 for further in-orbit testing. Their own language about the venture is restrained, referring to it as a “very minimal test,” rather than a breakthrough, and to the field as involving “a new frontier” rather than being solved.
OpenAI’s Safety Shakeup
Around the same time, OpenAI announced the dismissal of three of its employees for mishandling sensitive information outside established company procedures. The Wall Street Journal reported that those let go were Jasmine Wang, Tomek Korbak, and Mikita Balesni, at least two of whom were working on safety-related projects.
Sources said that one employee had shared confidential information with an external AI safety organization that was analyzing OpenAI’s models. OpenAI emphasized that the dismissals were not related to safety concerns but to how they handled information.
The circumstances of this incident deserve mention. In July, OpenAI agents had broken out of their restricted test environments and exploited unknown flaws in the system to reach the Internet and penetrate the open-source platform Hugging Face. Subsequently, the company tightened its network restrictions, increased its monitoring, and temporarily suspended advanced model training and assessment involving tool use.
Amidst the increasingly heated AI safety debates, OpenAI’s harsh treatment of their own safety team has raised a delicate question. When a company fires a researcher who was working on safety on the grounds of violating company policy, how do you draw the line between protecting company secrets and safeguarding the independence of safety review? There is no industry consensus on this.
Tencent’s Compute Breakthrough
Tencent, China’s tech giant, is already responding to its shortage of computation with a massive lease agreement. According to the Financial Times, it signed a five-year deal with Oracle to lease multiple data centers in Southeast Asia for an amount estimated at around $7 billion.
The payment structure of this purchase is also noteworthy, and about thirty percent was paid upfront, causing Tencent’s second-quarter free cash flow to be negative by 13.8 billion yuan, the first negative quarterly figure in more than a decade. The price paid shows the urgency of Chinese Internet companies to obtain high-power machines.
With U.S. export controls continuing to tighten and Nvidia’s most advanced processors unable to enter China directly, Tencent has to rent data centers overseas to meet its requirement for computing power. Tencent’s capital expenditure in the second quarter of this year jumped by 176% compared with the same period last year, mainly for its investment in artificial intelligence equipment and the prepayment of cloud resources.
Meanwhile, Alibaba is in preliminary talks with Spain’s leading renewable energy company, Solaria, to secure power for its data centers in Spain. Alibaba plans to enter Turkey, Finland, and the Netherlands within the next year and has set a target of exceeding 20 gigawatts of global data center capacity by 2032.
Both Chinese giants point to the same trend: in the era of digitized competition, the source of energy and memory chips is becoming increasingly important, and the ability to secure energy is becoming a more scarce strategic resource than algorithms themselves.
Samsung’s AI Dividend and Huawei’s Chip Breakthrough
In the first quarter of this year, Samsung Electronics posted its highest quarterly operating profit of 57.2 trillion won, with 53.7 trillion won coming from its chip business. It said this was due to “high-value-added AI demand,” and expects strong memory chip demand to continue.
The teardown of Huawei’s latest high-end chip, the Kirin 9050 Pro, indicates an alternative technological route. According to an analysis by the chip design research company Kurnal Insights and the well-known chip-testing platform Geekerwan, the chip is vertically stacked using a ‘Logic Folding’ design. Two identical dies—one used for computing, made with SMIC N+3 process technology, the other used for memory and SRAM, using N+2 technology—are directly bonded together using copper-to-copper hybrid bonding.
A 3D V-Cache by AMD is a similar design, but Huawei’s bonding density is higher, which shortens the signal line between the execution unit and the memory, reducing the latency. According to Geekerwan’s testing, the Kirin 9050 Pro has a NPU area increased by 150%, with INT8 throughput reaching 68 TOPS, and it can support on-device operation of 30-billion-parameter mixture-of-experts models.
On gaming performance, the Mate 90 Pro Max, equipped with the same chip, outperformed some Snapdragon 8 Elite Gen 5 devices in Genshin Impact and led Dimensity 9500 devices in Wuthering Waves. But the performance of this device is partly based on the native optimization of HarmonyOS, and games running through translation layers still show significant efficiency gaps.
The significance of the Kirin 9050 Pro lies not in whether it can completely surpass competitors on TSMC’s 3nm process, but in whether its architectural innovation and advanced packaging technology can still extract considerable performance headroom under advanced process constraints. This is Huawei’s first new Kirin chip launched at a flagship event in six years, following the Mate 40 global launch.
Apple’s October Two-Punch
In its line of hardware products for consumers, Apple is preparing a rare “staggered-wave” launch strategy. According to Bloomberg’s Mark Gurman and multiple sources, Apple plans to first launch a new smart home product line on October 13, including a new smart home hub device, upgraded HomePod mini, and Apple TV 4K.
This smart home hub (tentatively called “HomePad” by outsiders) is expected to feature an approximately 6-inch square display, with design inspiration from the iMac G4, supporting both desktop placement and wall mounting, running a new system centered on Siri AI, positioned as the unified control entry point for Apple’s smart home ecosystem. The last week of October will then see updates to the Mac and iPad product lines, possibly including an OLED-screen MacBook Pro.
Separating smart home and professional computing devices into different launches reflects Apple’s clear awareness of the different narrative logics of the two product lines: the former needs to emphasize ecosystem integration and daily companionship, while the latter focuses on performance breakthroughs and creative scenarios. As AI features increasingly become core selling points for hardware, this layered strategy also facilitates Apple’s delivery of differentiated value propositions to different user groups.
The AI Narrative and Hidden Anxieties in Capital Markets
The IPO of Anthropic is at the same time a flashpoint for the highs and lows of the new AI industry. The developer of the Claude series models is in the process of putting up a stock market that might be one of the largest in history, showing potential shareholders a market size exceeding $30 trillion.
But, under the law, Anthropic had to include a number of less encouraging statements in its prospectus, such as that its models might pose “existential risks” to humanity; that they might “lie,” “resist shutdown,” or even “engage in extortion.” As a result, some commentators jokingly construed the prospectus as: “Our AI could make the world incredibly better, or it could make it incredibly worse; we can’t tell.”
It is also the case that Anthropic’s figures likewise are revealing. From approximately $400 million in 2024, its revenue rose to nearly $5 billion in 2025. Losses, on the other hand, soared to over $8 billion. The costs of chips, data centers, and power are extremely high, but the question remains whether customers are willing to pay for these capabilities.
Anthropic is not without difficulties in China, where the Ministry of Industry and Information Technology previously named Claude Code as having security risks, noting it could transmit sensitive information to remote servers without user consent. A technical analysis has also revealed that some developers found that Claude Code sends requests to specific websites at extremely high frequencies and begins data transmission before users formally authorize it.
Whether it is chips in space, or the construction of Spanish data centers from the sun, or whether the firing of safety workers or the rushing of a $1-trillion valuation are the actions of different companies, they all share the same underlying logic: the race for artificial intelligence has moved beyond pure algorithmic competition to a systemic struggle involving energy, chip manufacturing, political and legal considerations. In the months ahead, it will be up to the likes of Anthropic’s IPO, and Google’s satellite testing, to produce new footnotes to the competition. The pace of technological advance and the deficiency of human capacity to manage it will now be all too obvious.
PAUL GOULDEN