Bitget App
Trade smarter
Open
HomepageSign up
Bitget>
News>
Meta (META.US) strengthens self-developed AI chips! Plans to deploy data centers in the first half of next year, aiming to reduce inference costs and energy consumption

Meta (META.US) strengthens self-developed AI chips! Plans to deploy data centers in the first half of next year, aiming to reduce inference costs and energy consumption

智通财经2026/09/15 16:16
By: 智通财经
Meta Platforms plans to deploy its next-generation self-developed artificial intelligence chips to data centers in the first half of 2027, aiming to reduce the energy consumption and costs required to run AI models through customized chips.

According to Zhitong Finance APP, Meta Platforms (META.US) plans to deploy its next-generation proprietary artificial intelligence chips to data centers in the first half of 2027, hoping to reduce the energy consumption and cost required to run AI models through custom chips. Meanwhile, the company has committed to a deployment scale exceeding 1 gigawatt for the relevant chips, and indicated that if AI demand remains strong, the subsequent deployment speed will further accelerate.

Yee Jiun Song, Meta's Vice President of Engineering, stated in an interview that the company's third-generation self-developed AI processor, MTIA 450, is currently in the testing phase, with the chip code-named "Arke"; the next generation MTIA 500 is code-named "Astrid" and is expected to complete design in about a month, with plans to enter data centers before the end of 2027.

Song noted that each generation of Meta's custom chips takes on greater technical challenges in exchange for higher performance, including improved performance-per-watt and performance-per-cost. This also means that as investments in AI infrastructure continue to expand, Meta is trying to improve computational efficiency and reduce long-term operating costs through in-house chip development.

Partnering with Broadcom for Design, Manufactured by TSMC, Deployment Scale to Exceed 1 Gigawatt

Meta is currently cooperating with Broadcom (AVGO.US) to develop custom AI chips, with TSMC (TSM.US) responsible for manufacturing. Song revealed that Meta has committed to deploying over 1 gigawatt of related chips within 12 months. If AI demand remains strong, the company expects deployment speed to accelerate further.

Currently, Arke has entered the practical testing phase. On September 1, the first batch of 12 Arke chips was delivered to Meta by TSMC, with actual performance differing from previous simulation results by only 2% to 3%.

More notably, these processors have already successfully run Meta's own AI models, as well as models from DeepSeek and Alibaba (BABA.US), showing that Meta's proprietary chips are not only optimized for a single internal company model but are capable of running various AI models.

Abandoning the Dual Focus on Training and Inference, Meta Shifts Proprietary Chip Emphasis to AI Inference

Meta's self-developed AI chip strategy has also undergone adjustments.

The company previously planned to develop a chip called Olympus for both AI model training and inference, but later canceled the project in part for cost considerations. Since then, Meta has increasingly focused its proprietary chips on AI inference. Song stated, "These will become our main chips for general-purpose inference."

As Meta continues to integrate AI features into Facebook, Instagram, and other products, the frequency of model calls keeps increasing, which in turn raises the computational resources needed for inference. Therefore, rather than simply pursuing greater computing power, reducing the cost and energy consumption of each AI inference is becoming increasingly important for Meta.

As a result, Meta's chip strategy is becoming clearer: rather than attempting to cover all AI computational tasks immediately, the goal is to first build specialized chips for large-scale, ongoing inference workloads, thereby lowering the overall operational cost of AI infrastructure through higher performance-per-watt and performance-per-dollar.

AI Infrastructure Race Extends to Proprietary Chips, Meta Claims to Have a “Very Robust” Product Roadmap

As major tech companies continue to expand their investment in AI infrastructure, proprietary chips are becoming a crucial way to control AI costs and enhance data center efficiency. Meta's progress with MTIA 450 and MTIA 500 shows that its in-house chip project is moving from testing to large-scale deployment.

According to current plans, the MTIA 450 now being tested will enter data centers in the first half of 2027, while the next-generation MTIA 500 is expected to be deployed before the end of 2027. Future product generations will further enhance speed and throughput. Song stated that Meta has established a “very robust roadmap” for its custom chips.

Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
Understand the market, then trade.
Bitget offers one-stop trading for cryptocurrencies, stocks, and gold.
Trade now!

You may also like

Altcoin Breakout Tests Long-Term Resistance
Cryptonewsland2026/09/15 16:51

Trending news

More
1
Altcoin Breakout Tests Long-Term Resistance
2
Dow Jones Industrial Average slides as yields beat the Fed to tomorrow's hike

Crypto prices

More
Bitcoin
Bitcoin
BTC
$76,434.83
-3.03%
Ethereum
Ethereum
ETH
$2,423.69
-4.37%
Tether USDt
Tether USDt
USDT
$0.9994
-0.04%
BNB
BNB
BNB
$718.25
-0.74%
XRP
XRP
XRP
$1.39
-2.22%
USDC
USDC
USDC
$0.9999
-0.01%
Solana
Solana
SOL
$99.19
-3.41%
TRON
TRON
TRX
$0.3356
-1.78%
Hyperliquid
Hyperliquid
HYPE
$77.18
-4.55%
Zcash
Zcash
ZEC
$1,120.36
-2.18%
How to buy BTC
Bitget lists BTC – Buy or sell BTC quickly on Bitget!
Trade now
Become a trader now?A welcome pack worth 6200 USDT for new users!
Sign up now
Trade smarter