Today’s rundown:

• Microsoft has Plans for Wireless Data Centers…

• It’s Official: There’s a Data Center on the Moon, Kind of

• Deals and Dev: Amkor advancing $2B Phoenix facility, Blackstone to swallow construction firm, and Together AI seeks $100M

• More to Explore: Nvidia’s blog explores AI, Nuclear fission machine smashes records, Cerebras donates exaflops for the national cause

Est. read time: 6mins 27secs

Microsoft Wants to Take its Data Centers Wireless

Microsoft is exploring innovative approaches to enhance data center connectivity, particularly through the utilization of high-frequency wireless communications.

Going Sub-terahertz

The company is considering the deployment of frequencies in the 246GHz-275GHz range, which is earmarked for potential future 6G networks. This high-band spectrum could facilitate the transmission of substantial data volumes, over short distances due to the inherent physical limitations of high-frequency signals.

The rationale behind Microsoft's interest in this "sub-terahertz" (sub-THz) spectrum for data center applications is multifaceted.

• The potential for highly directional antenna arrays to create multiple concurrent communication links via spatial multiplexing (see video below). This is made feasible by the short range of sub-THz frequencies, which also promotes spatial reuse by minimizing interference.

• Additionally, the use of directional antennas could allow for the flexible establishment and disconnection of radio frequency links as needed.

Moreover, the controlled indoor environment of data centers, coupled with potential enhancements to building infrastructure, could support the creation of high-bandwidth, high-data-rate links. This setup would minimize interference with external deployments, thereby fostering efficient spectrum sharing and coexistence.

Faster Connections, No Wires

Microsoft is also considering the integration of free-space optical (FSO) technology, which transmits data via light in open space rather than through traditional cables.
However, the company acknowledges challenges with FSO, particularly in maintaining precise alignment of optical beams amidst equipment vibrations. In contrast, the broader and adjustable beam width of sub-THz RF technology is less susceptible to such alignment issues.

The capacity for rapid electronic steering of RF beams is seen as a way to mitigate vibration-related challenges, enabling the development of control mechanisms to ensure stable and reliable wireless communication within data centers.

More Context

From the exclusive at LightWave, "Following a torrid pace of leasing in 2023, Microsoft continues to be the most active leaser of data center capacity amongst the hyperscalers," wrote the financial analysts at TD Cowen in a recent note to investors, comparing Microsoft against other hyperscale companies like Amazon, Meta and Google.
"Furthermore, our checks indicate that the driver behind the [Microsoft's] heavy data center capex investment is its internal view that there is a clear pathway to monetizing AI via the integration of AI into both its existing product set and the launch of new AI products. This, coupled with its selection of an AI deployment architecture ahead of its hyperscale peers, has given Microsoft a notable head start in securing [data center] capacity vs. peers."

Data Center on the Moon… Kind of

The outer boundary of the data center frontier has expanded to the lunar surface, thanks to a small data storage device from Lonestar Data Holdings.

Lunar Data Center

This innovative venture became a reality when Lonestar Data, a customer on a lunar lander from Intuitive Machines, successfully landed on the moon. Despite initial communication hurdles, the lander is now upright and actively transmitting data back to Earth.

The Intuitive Machines Odysseus lander's successful touchdown heralds the first U.S. lunar landing since Apollo 17 in 1972, under NASA’s Commercial Lunar Payload Services (CLPS) initiative. This mission aims to foster a lunar technology ecosystem, paving the way for future manned missions.

For Lonestar Data, it serves as a proof-of-concept for their vision to provide off-planet disaster recovery services, with ambitions to develop sophisticated data centers on and around the Moon.

The successful transmission of the U.S. Declaration of Independence to the lunar lander demonstrated Lonestar's capability for off-planet data storage and recovery. This test is a precursor to more ambitious plans, including storing data for the State of Florida on a larger data storage unit to be sent to the moon later this year.

Lonestar Data stands at the forefront of off-planet computing, aiming to offer disaster recovery as a service (DRaaS) from the moon. The company envisions providing secure digital services for enterprise and government customers, in partnership with cloud platforms, with a goal to launch commercial services by 2026.

CEO Chris Stott says the company is “one step closer to realizing our vision of establishing a secure, off-planet data storage infrastructure that can serve humanity for generations to come.”

Lonestar's success contributes to NASA’s broader vision of a lunar economy supporting the Artemis project, which seeks to establish a long-term human presence on the moon.

The concept of lunar data centers, once a topic of space conferences, is now a reality, propelled by advancements in space technology and the advent of private space missions. As Lonestar Data Holdings charts new territory with its lunar data storage, it underscores the expanding role of data security and storage in the era of lunar exploration and beyond, marking a new chapter in the space revolution.

- Deals and Developments -

Amkor Pushing Full Speed Ahead on $2B TSMC FAB

Amkor is moving full speed ahead with its $2 billion TSCM-located advanced packaging facility.

Advanced packaging refers to the costly, technical process of combining multiple semiconductor devices into a singular device.
This is usually done in Asia, so it's a big deal vis-à-vis our reliance on the Asian market for our semiconductors.

The Amkor facility will boast 500,000 sq ft of cleanroom space dedicated to high-performance computing, automotive, and communications markets. Analysts predict the development will create as many as 2000 local jobs.

Amkor is closely working with the CHIPS Program Office, optimizing designs, and gearing up for high-volume manufacturing in 2-3 years.

Despite a challenging environment, Amkor's Q4 earnings surpassed expectations, showcasing resilience and strategic growth in key sectors like premium smartphones and electric vehicles.

Read the full write-up here.

Blackstone Looks to Acquire European DC Construction Firm

Investment giant Blackstone is reportedly in the lead to acquire Winthrop Technologies, an Irish specialist in data center construction, with the deal speculated to be worth approximately $878 million. The investment giant is looking to secure a majority stake in Winthrop

Founded in 1995, Winthrop initially operated as a conventional mechanical and electrical contractor across various sectors but has shifted its focus towards mission-critical and turnkey data center solutions.

Boasting a workforce of more than 750 employees, Winthrop is currently constructing several large-scale data center projects across seven European countries, representing over 430MW of capacity. The company's financial performance looks good, with revenues exceeding $735 million in 2021 and project deliveries worth more than$1 billion in 2022.

In the past two years, Blackstone has acquired or partnered with digital infrastructure firms like:

• QTS Realty Trust

• Digital Realty

• CoreWeave

• Phoenix Tower Int.

• Therma Holdings

• RE Tech Advisors

Here’s Data Center Dynamics’ full article on the acquisition.

Better Together: AI Platform Looks to Raise $100M

Together Computer Inc., a startup operating a cloud platform for artificial intelligence workloads, is reportedly seeking to raise over $100 million.

The round is expected to value the San Francisco-based company at over $1 billion, doubling its valuation since its last funding round in November, which included Nvidia Corp. among its investors.

Together's platform offers access to Nvidia's H100 and A100 GPUs, allowing customers to deploy scalable GPU clusters, simplifying the creation of AI infrastructure.
The company also provides software features to aid in training large language models, including an open-source optimization tool called FlashAttention-2, which enhances the use of SRAM in GPUs for improved performance.

Additionally, Together offers a training dataset called RedPajama-Data-v2, comprising over 100 billion documents in five languages, designed to streamline the filtering process for AI projects.

The funding is intended to support product development, with the cost of a single H100 graphics card reported to be between $25,000 and $40,000.

Together faces competition from other AI-optimized cloud platforms, such as Lambda Labs which recently closed a $320 million round at a $1.5 billion valuation, and CoreWeave, which secured $2.3 billion in debt financing from Coatue and other institutional investors.

Here’s SiliconAngle’s article about the raise.

- More to Explore -

1. From the Invidia Blog: Nvidia does a deep dive into how AI is shaping up to affect modern computing. Discover how industries and governments are using AI and accelerated computing to drive innovation, enhance efficiency, and reduce environmental impact, transforming the way we live and work.
Dive in with the chip GOATS on AI here.

2. Nuclear Fusion Machine Smashes Records: Gizmodo highlights the Joint European Torus, a UK-based nuclear fusion project, has achieved a breakthrough by generating a record 69 megajoules of heat, marking a significant stride towards cleaner, unlimited energy.

Read Gizmodo’s piece here.

3. 4 ExaFlops for the National Cause? Cerebras announced recently that it will support the U.S. National Science Foundation’s (NSF) National Artificial Intelligence Research Resource (NAIRR) Pilot with a remote access grant of up to 4 exaFLOPs of AI compute for large language and generative AI model training using CS-2 systems.
The full press release can be found here.

That’s it. Thanks for reading.

Let me know what you think by replying to this email. Also, sharing it with someone who loves data centers helps too.

- Taylor