Tech Express | October 13, 2026 — DeepSeek Doubles Its Round to About RMB 100 Billion

1. DeepSeek Doubles Financing Round to About RMB 100 Billion

Source: Mydrivers

DeepSeek is set to close its latest funding round at roughly RMB 100 billion, up from the RMB 50 billion originally planned, according to people involved in the deal cited by media reports. It is the company's second round, with new and existing shareholders each contributing about half. The larger size did not change its valuation; the pre-money valuation remains around RMB 500 billion.

New investors include Shixi Capital, Zhongxin Juyuan (SMIC Capital), CPE Yuanfeng, Legend Capital and Geely, mostly industrial investors. Existing backers from the first round — CATL, Tencent, IDG Capital, Lishi Capital and Shixiang Capital — are also on the list. CATL plans to add about RMB 10 billion, after contributing about RMB 5 billion in the first round. If completed, DeepSeek will have raised about RMB 150 billion in half a year — the largest primary-market financing in China this year.

Media reports say most of the proceeds will go into computing infrastructure. DeepSeek is building a large data center in Inner Mongolia expected to deploy at least 160,000 Huawei Ascend 950-series chips, with a chip procurement budget alone of about RMB 17 billion. Founder Liang Wenfeng has also been focusing on embodied AI, and DeepSeek plans to lead an investment in robotics startup Winverse AI.

Source

2. Developer Rebuilds Call of Duty: Modern Warfare 2 Source Code With AI Agents

Source: Mydrivers

Developer Maurice Heumann has detailed in a blog post how he used AI agents to rebuild the source code of Call of Duty: Modern Warfare 2 (2009). The project took three months and consumed more than 500 billion tokens. About 99% of functions have been reproduced, and 83% match the original byte for byte; the game reportedly runs perfectly.

Heumann used agents from both Anthropic's Claude and OpenAI's Codex. He started with four agents and had scaled to 17 working in parallel by September 30. The work relied mainly on Sonnet 5, along with a good deal of Opus 5.5 and OpenAI's Luna, Sol and Terra. He estimates the project actually consumed 600 to 700 billion tokens, costing around EUR 1,200 under a subscription model, versus USD 200,000 to 250,000 if billed by API.

Heumann said the biggest challenge was not generating code but verifying correctness. In the first month the code was highly readable yet semantically wrong: function signatures, types and struct layouts were off, and agents made unnecessary architectural changes. The team moved to automated verification, switching to the compiler used by the original game and comparing every rebuilt function byte by byte, with only pass or fail as outcomes. Once the verification script was in place, the models began to work properly and costs fell. Heumann also said Activision has applied copyright pressure, and his two earlier blog posts were removed.

Source

3. Qwen to Launch a 'Qwen Go' Personal Agent That Can Win Compensation

Source: Mydrivers

Qwen's Personal Agent, named 'Qwen Go!', is about to launch, according to reports. One user who remembered a flight had been delayed but could not recall which one asked Qwen Go to pursue compensation. The agent first checked the booking platform's orders and flight status to confirm the delay, then contacted customer service. Faced with a voucher offer, Qwen Go insisted on cash compensation and ultimately won the user RMB 70.

Another user asked Qwen Go about Taobao's price-protection policy, saying they had heard they could get money back if an item got cheaper but did not know how. Qwen Go found the price-protection entry on Taobao, checked all the user's orders and automatically identified an eligible order for a non-stick wok, refunding RMB 52 in price difference.

Qwen is described as the world's first personal AI assistant that can get things done for users; since launching last year it has gained capabilities in shopping, ride-hailing, delivery and finance through Alibaba's ecosystem and open platform. In videos shared by users, the upgraded Qwen can query and act on various authorized platforms, compare options and call service providers to complete tasks.

Source

4. Former Plaud Executive Found Ambi, Raises $26M for an Always-Listening AI Assistant

Source: TechCrunch

Former Plaud executive Peter Mo is building Ambi, an AI app designed to give users real-time intelligence and insights by listening to them continuously, TechCrunch reported. The company said on Monday it had raised $26 million across multiple funding rounds led by Luminous Ventures, a backer of Meituan and Unitree Robotics, with original design manufacturer Huaqin also participating.

Ambi is also releasing apps for iPhone, Mac and Apple Watch. The apps work like meeting notetakers, but users can let them listen so they capture context all day. The company promises its AI agent can proactively offer suggestions or surface relevant knowledge during a meeting, based on a user's calendar, meetings, knowledge base and web data, and that users can start a task on one device and continue it on another.

Peter Mo said the way people interact with intelligence is limited, and that intelligence is often trapped on the screen; with hardware such as Plaud, users must wait for a recording to transfer to the phone and upload to the cloud before they get intelligence. That is why, he said, Ambi aims to make AI usable anytime and anywhere. The company also plans for its upcoming desktop app to observe users' work and store important parts in memory for more personalized suggestions.

Source

5. IDC: Nvidia's China AI Chip Share Falls to 34% as Huawei Narrows Gap to 28%

Source: Mydrivers

Nvidia's share of China's data-center AI chip market fell to 34% in the first half of this year, with Huawei close behind at 28%, narrowing the gap to 6 percentage points, according to an IDC report. Nvidia shipped 678,000 AI accelerators in China from January to June, down 44% year on year, as US export controls limit supply of its most powerful AI chips to China and the country pushes domestic alternatives.

Among domestic vendors, Huawei's AI accelerator shipments rose from more than 300,000 in the same period last year to 572,000, giving it a 28% share; T-Head took 7%, Cambricon 5%, Kunlunxin 4% and Hygon 3%. Nine of the 11 Chinese chipmakers tracked by IDC posted year-on-year shipment growth. Nvidia CEO Jensen Huang has repeatedly criticized the policy, saying a USD 50 billion Chinese AI chip market is being handed to Huawei.

Total AI accelerator shipments in China reached 2.008 million units in the first half, of which GPU shipments fell 26% to 991,000 while non-GPU accelerator shipments grew nearly 80% to 1.016 million, surpassing GPUs for the first time. IDC also expects total AI accelerator shipments in China to grow 19% this year, while GPU shipments fall 9%.

Source

6. Memory Prices Swing Wildly as DDR4 Contracts Jump 24% in Three Months

Source: Mydrivers

Global memory prices are undergoing sharp swings, with DDR4 contract prices up 24% over the past three months, outpacing the more advanced DDR5, which rose 20% over the same period, according to a new report from investment bank Barclays. As of October, DDR4 chip contract prices were up 4% month on month and 24% over three months; DDR5 contract prices rose 3% and 20% respectively over the same spans.

Barclays noted that a persistent memory shortage, combined with long-term agreements, has brought predictability to the memory chip business, with supply-agreement prices now closer to market pricing and traditional price ceilings and floors fading. The findings were echoed by a separate Citi analysis.

Citi expects the DRAM and NAND supply gaps to persist until 2028: the DRAM gap is projected to peak at 13.9% in the fourth quarter of 2027 and the NAND gap at 8.3% in the first quarter of 2028. Because AI-focused HBM also depends on DRAM capacity, manufacturers have shifted capacity toward DRAM, which in turn tightens NAND supply and creates a chain reaction.

Source

7. Apple Confirms an October 13 Event as Three Products Focus on the Smart Home

Source: 9to5Mac

Apple has officially confirmed a product launch event on October 13. According to 9to5Mac, the event carries the tagline 'Welcome home' and is expected to focus on new smart home products. It has two parts: an in-person experience for press in New York City and what Mark Gurman has called an 'online video presentation'. Press releases and video are expected around 9 a.m. ET on October 13.

The headlining product is expected to be Apple's long-rumored smart display, which Bloomberg says will be called HomeView. The device reportedly has a 6-inch square display atop a half-dome base held up by a metal arm, with wall-mounted and countertop versions, and a design likened to Apple's iconic iMac G4. One flagship feature is its ability to recognize multiple people in a household, telling who is speaking or approaching and adjusting what is shown on screen. HomeView is not expected to be available until November, however.

Apple is also expected to announce a new Apple TV 4K, the set-top box's first update since November 2022. The new Apple TV will look the same as its predecessor but is reported to include an A19 or A19 Pro chip, bringing Siri AI and Apple Intelligence features to Apple TV for the first time. It may also add built-in microphones for Siri interactions, while the Siri Remote could gain a speaker to make it easier to find.

Source

8. Google Launches the Fitbit Edge Band With a 46% Larger Screen and a Replaceable Battery

Source: Mydrivers

Google has officially launched its new Fitbit Edge band, now available for pre-order and shipping October 27. The US price is USD 179.99 (about RMB 1,205), with a Stephen Curry special edition featuring a custom band at USD 219.99 (about RMB 1,472). Google Health teased the product on social media on October 9 with the line 'Get an Edge on your health. 10.12.2026', and it arrived three days later.

Positioned between the Pixel Watch and the screenless Fitbit Air, the Fitbit Edge is a screen-equipped band that Bloomberg reports will replace the Charge 6, with the Charge line discontinued. Its 1.34-inch AMOLED screen has 46% more display area than the Charge 6 and is covered with Gorilla Glass 3. The body uses a recycled aluminum case, is 11.5 mm thick, supports IP68 and 5ATM water resistance, keeps a physical button on the right, and comes in four colorways.

The Fitbit Edge has built-in GPS, which Google says is 40% more accurate than the Charge 6, plus an altimeter, blood-oxygen and skin-temperature monitoring, and support for the AI health coach Google Health Coach, which can broadcast heart rate to gym equipment. Battery life is about seven days with the always-on display off and about two days with it on, with five minutes of charging providing a day of use and the battery replaceable. It runs a Fitbit system compatible with Android and iOS and includes three months of Google Health Premium.

Source

9. Google's Space AI Debut Sends Four TPU Chips Into Orbit

Source: Mydrivers

Google's Project Suncatcher space AI initiative has taken a key step: its first experimental prototype satellite reached orbit aboard SpaceX's Transporter-18 mission, and the ground team has made contact, with operations matching expectations. The satellite, developed with Planet Labs, is about the size of a refrigerator and carries four of Google's own TPU chips.

Google says it is not a commercial space data center but a way to validate the chips' physical durability in the real space environment. Over the coming weeks it will collect on-orbit data, focusing on how the TPUs perform under launch vibration, space radiation and sharp temperature swings. Because space is near-vacuum, heat cannot be removed by air convection; it must travel through heat pipes to a radiator and be radiated away. That directly limits continuous runtime: the TPUs can compute for only about 15 minutes at a time before they must pause to cool.

The satellite will also run Google's open-weight Gemma AI model. Google's long-term vision is to cluster multiple TPU-carrying satellites in orbit, linked by high-speed laser communications, and it plans to launch two more satellites in 2027 to test high-bandwidth laser links. Cost remains a major obstacle: Google estimates that for a space data center to be cost-effective, SpaceX's Starship would need to make about 1,800 launches over 10 years.

Source

10. IDC: China Smartphone Shipments Fall 15.4% in Q3 as Huawei Grows

Source: Mydrivers

China's smartphone shipments came to about 57.9 million units in the third quarter of 2026, down 15.4% year on year, according to preliminary data from IDC's quarterly mobile phone tracker. The decline accelerated markedly from roughly 4% in the first half, as the market continued to come under pressure.

Notably, Huawei was the only top vendor to post double-digit year-on-year shipment growth, reaching a 21.1% share. IDC said Huawei extended its strong run this year with a diversified product lineup and relatively restrained price increases. Apple's new iPhone saw a better-than-expected reception, but with fewer models launched this year its quarterly shipments still slipped slightly year on year; Honor was the only major vendor to grow sequentially.

IDC attributes the overall decline to sharply rising costs for memory and other components, which are pushing handset prices higher and forcing some vendors to adjust product specs, further dampening upgrade demand. IDC expects memory and other material costs to stay high for several more quarters, with phone prices likely to keep rising and cost pressure possibly lasting into 2027.

Source