On-Prem

Tencent shrugs off effect of US sanctions and says Chinese suppliers will fill GPU gap

Ordered enough Nvidia kit to keep evolving its AI and has plans to work around bans


Tencent placed early, and large, orders for Nvidia GPUs, believes it procured enough of the hardware to evolve its AI for another two generations, and has plans to keep improving performance – even if US chip bans put GPUs off limits for the foreseeable future.

So said the Chinese web giant's president, Martin Lau, on Tencent's Q3 earnings call this week.

"In terms of the chip situation right now we have one of the largest inventories of AI chips in China," Lau told investors. "We were the first to put in an order for the H800" – a card Nvidia designed specifically to sit below the threshold of the US's tech export bans. Subsequent bans forbade future sales of the H800.

Tencent doesn't much care.

"We have enough chips to continue development for at least a couple of generations" Lau revealed, referring to future evolution of Tencent's "Hunyuan" proprietary large foundation model.

Investors were told Hunyuan is already summarizing meetings staged on Tencent's Zoom-like Meeting platform, offering productivity-enhancing suggestions to Tencent's coders, and improving the efficiency of art creation at its game studios.

The AI is also writing copy for advertisers on Tencent's properties, targeting those ads more precisely, and improving click-through rates. Performance at Tencent's video ad business has improved as a result.

In the near future, Lau expects Hunyuan to write ads and respond to customer inquiries.

"A lot of merchants can combine advertising and sales processes into one," Lau enthused, adding that Tencent envisages a customer-facing AI – he used the term "smart agent" – that will see the megacorp's AI take on an even broader role for its business customers.

That's doable, he opined, without needing to buy more GPUs.

"The ban does not affect the development of Hunyuan and our AI capability," Lau declared.

Bans may slow Tencent a little, though, as Lau admitted "Going forward we have to figure out how to make usage of our AI chips more efficient." He then mused: "Can we offload inference to lower performance chips, and preserve high performance chips for training?"

Tencent plans to look into it.

And when it does need more silicon for training, Lau said Tencent will look for a Chinese supplier.

Which isn't supposed to exist. US sanctions are designed in part to stop China developing equivalents to the H800.

Lau appears to believe China's semiconductor industry will rise to the occasion.

Which, if true, leaves the US policy looking a bit meek. The readout of the meeting between presidents Biden and Xi included the following:

The president emphasized that the United States will continue to take necessary actions to prevent advanced US technologies from being used to undermine our own national security, without unduly limiting trade and investment.

It appears that Tencent, for one, thinks China will soon have its own tech at a level of sophistication the US thinks threatens its security.

The Chinese web giant posted Q3 revenue of $21.5 billion – a ten percent year-on-year improvement. Operating profit fell six percent, but still came in at $6.8 billion.

The firm highlighted improvement to its business services division, helped by its past decision to offload loss-making IaaS clients from the Tencent cloud.

WeChat and QQ, Tencent's social platforms, won 1.336 billion and 558 million monthly active users apiece – each up slightly year on year.

CEO and chair "Pony" Ma Huateng said the results vindicated Tencent's move to "high quality revenue growth" fuelled by its video and gaming services. ®

Send us news
Post a comment

We heard you like HBM – Nvidia's Blackwell Ultra GPUs will have 288 GB of it

There goes AMD's capacity advantage

Nvidia's Vera Rubin CPU, GPU roadmap charts course for hot-hot-hot 600 kW racks

Now that's what we call dense floating-point compute

Nvidia wants to put a GB300 Superchip on your desk with DGX Station, Spark PCs

Or a 96 GB RTX PRO in your desktop or server

A closer look at Dynamo, Nvidia's 'operating system' for AI inference

GPU goliath claims tech can boost throughput by 2x for Hopper, up to 30x for Blackwell

Dell discloses monster 20-petaFLOPS desktop built on Nvidia's GB300 Superchip

HPE and Lenovo also have plans to put Jensen's latest hardware to work

Nvidia punts silicon photonic switches to keep GPUs fed with data

Power sipping bandwidth bottleneck busters – or that's the hope, anyway

Nvidia invests in quantum computing weeks after CEO said it's decades from being useful

Qubit awkward, you might say

Tencent slows pace of GPU rollout as DeepSeek helps it wring more performance from fewer accelerators

Chinese giant says locals are more efficient than Western hyperscalers, and has tiny capex to prove it

Nvidia won the AI training race, but inference is still anyone's game

When it's all abstracted by an API endpoint, do you even care what's behind the curtain?

Schneider Electric plugs into AI's power hunger with Nvidia digital twin tech

Because guesswork won't keep the lights on

FCC on the prowl for Huawei and other blocked Chinese makers in America

Be vewy vewy quiet, I'm hunting rackets

China bans compulsory facial recognition and its use in private spaces like hotel rooms

PLUS: Zoho's Ulaa anointed India’s most patriotic browser; Typhoon-like gang targets Taiwan; Japan debates offensive cyber-ops; and more