Capital · United States

Reflection AI debuts open-source Beam model with 501B parameters

Credited to SiliconANGLE · siliconangle.com · 2d ago

Photograph · SiliconANGLE

Reflection AI Inc. today introduced Beam, an open-source large language model with 501 billion parameters.

The launch comes a few months after the startup raised funding at a $25 billion valuation. Around the same time, Reflection AI reportedly inked a $6.3 billion deal with SpaceX Corp. to rent Nvidia GB300 NVL72 appliances. It used those systems, which each contain 72 graphics cards, to train Beam.

Reflection AI benchmarked its model against GLM-5.2, an open-source LLM that includes about 250 billion more parameters. The company determined that Beam can perform some tasks better while using between one third and one fourth the hardware. Furthermore, it says that the LLM approaches the performance of Qwen 3.8-Max, a model with over 2 trillion parameters.

The launch is particularly notable because many of the open-source ecosystem’s most advanced LLMs were created by Chinese companies. That includes Qwen 3.8-Max and GLM-5.2. Beam is the first open-source model from a U.S. startup to have demonstrated comparable or better performance. However, free LLMs continue to trail frontier models such as Anthropic PBC’s Claude Fable 5.1.

Reflection AI kicked off the development of Beam by training a relatively small prototype model. It then created a series of successively larger, more capable algorithms. The workflow eventually produced Beam Base, the foundation on which Beam is based.

The company developed Beam Base using a cluster of 6,144 graphics cards. It trained the model on 23.8 trillion tokens sourced from the public web and commercial sources. That dataset included a significant amount of software code. It created custom filters for each programming language to remove low-quality files.

Beam Base was developed in under four weeks. It subsequently performed midtraining, an optimization process that extended the model’s context window and enhanced its reasoning capabilities. That phase of the project set the stage for the third, most hardware-intensive phase of the training workflow.

Reflection AI spun up 10,000 GB300 graphics cards and used them to launch 1.3 billion reinforcement learning sandboxes. Those are virtual environments in which an AI model learns new skills. Beam’s sandboxes were optimized for tasks such as generating code, searching the web and running AI agents.

The company says the project’s reinforcement learning phase took only four weeks. It avoided unnecessary delays by developing software that prevented malfunctions from interrupting the workflow. It says its cluster achieved a median recovery time of eight minutes across the 71 errors that cropped up during training.

Beam is initially available through an early access program. Reflection AI plans to release its weights, documentation and fine-tuning tools later this month.

Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.

Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/

Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.

Atlassian lays groundwork for humans and AI agents to work side by side

Caterpillar and CoreWeave shorten the learning loop for physical AI

Google goes nuclear to power its AI data centers, committing to buy 890 megawatts of energy from Constellation

Meta teams up with Bret Taylor's Sierra Technologies on new standards for AI agent commerce

Multiply Labs raises $75M to optimize pharmaceutical manufacturing with robots

In the agentic era, standing privilege must go: Key thoughts from the SailPoint Navigate keynote

INFRA - BY MARIA DEUTSCHER. 10 HOURS AGO

SECURITY - BY ZEUS KERRAVALA. 10 HOURS AGO

Join our community

CapitalUnited States

Flex Secures $2B Investment for Axiom at $37.5B Enterprise Value

Credited to Converge Digest · convergedigest.com2d ago

Useful

Flex has secured a $2.0 billion convertible preferred equity investment in Axiom Solutions International, establishing an initial enterprise value of $37.5 billion for the AI infrastructure business that Flex plans to spin off as an independent public company in the first quarter of 2027. Funds affiliated with General Catalyst, Koch Equity Development and co-investors are purchasing the Series A Convertible Preferred Stock.

The financing provides equity capital for Axiom’s pending acquisition of EPC Power, the $4.4 billion transaction Flex announced in September to expand its grid-to-chip power portfolio. Flex has separately arranged committed term-loan financing for the remainder of the EPC Power acquisition. The preferred shares carry a 10% annual cash dividend before separation; afterward, 6% cash or 7% PIK.

The investment adds an external valuation benchmark to a business Flex has assembled around high-density AI infrastructure. Axiom will combine embedded and critical power, electrical infrastructure, power conversion, thermal management and compute integration. Flex continues to target completion of the Axiom separation in Q1 calendar 2027, subject to required approvals and customary conditions.

Investors — General Catalyst, Koch Equity Development and co-investors

The $37.5 billion enterprise value is the most consequential new data point in today’s announcement. When Flex formally named Axiom in September, the preliminary Form 10 established operating scale but did not provide this external investment valuation.

The financing clarifies how Flex intends to fund the next major expansion of Axiom. The pending $4.4 billion EPC Power acquisition moves the portfolio farther upstream into MW-scale power conversion, grid-forming controls, energy storage and emerging 800 VDC architectures.

The financing should not be treated as completion of either the EPC Power acquisition or the Axiom separation. Both remain subject to applicable approvals and closing conditions.

Company Profile • Axiom Solutions International

Flex Names $7B Infrastructure Spin-Off Axiom — Sept. 16, 2026

Flex Makes $4.4B Bet on 800V AI Data Center Power — Sept. 3, 2026

Flex to Acquire EP² for $1.1B — Mar. 30, 2026

Flex Debuts 33 kW Power Shelf for NVIDIA Blackwell — Aug. 5, 2025

Original · Converge Digest

Front Method

Capital · United States

AOI Raises $588M to Fund Manufacturing Expansion

Credited to Converge Digest · convergedigest.com · 2d ago

Photograph · Converge Digest

Applied Optoelectronics ( NASDAQ: AAOI ) has completed a previously announced $600 million at-the-market equity offering program, raising approximately $588 million in net proceeds as the optical component manufacturer expands production capacity for high-speed data center connectivity. AOI sold 5,694,845 common shares at an average price of $105.36 per share. The offering, first disclosed on August 21, represents the company’s third ATM equity program completed this year.

The new capital arrives as AOI scales manufacturing for 800G and 1.6T optical transceivers targeting hyperscale and AI data center networks. AOI said proceeds may be used for general corporate purposes including debt repayment, working capital and capital expenditures, with manufacturing expansion for data center optical devices identified as a primary focus. CEO Dr. Thompson Lin said the company plans investments in new production and R&D equipment as it expands advanced manufacturing capacity.

AOI’s manufacturing buildout is already substantial. In July, the company began developing nearly 400,000 additional square feet of manufacturing capacity at two properties in Pearland, Texas, focused on 800G and 1.6T transceivers. Earlier this year, AOI also secured a volume order exceeding $200 million for 1.6T transceivers from a major hyperscale customer. The company has been expanding capacity in both Texas and Taiwan as AI clusters drive higher optical bandwidth requirements.

Net proceeds — $587,999,985.89 after commissions and fees

Program — AOI’s third ATM equity program of 2026

Primary investment — Manufacturing expansion for data center optical devices

Other uses — Debt repayment, working capital, production equipment and R&D equipment

“We intend to use these proceeds to continue to make investments in the business, including new equipment and machinery for production and research and development use, as we continue to scale our advanced manufacturing capacity,” said Dr. Thompson Lin, Founder, President and CEO of AOI.

The significance of the transaction extends beyond the size of the equity raise. AOI is converting strong capital-market access into manufacturing capacity at the same time that its data center portfolio is moving from 400G toward much higher volumes of 800G and 1.6T products.

That relationship between capital and capacity is particularly visible in Texas. AOI began its Pearland expansion in July, adding nearly 400,000 square feet for high-speed transceiver manufacturing. Demand provides the other side of the equation: AOI disclosed in March that it had secured a 1.6T order worth more than $200 million from a long-standing hyperscale customer.

The $588 million infusion therefore gives AOI substantially more financial capacity to execute its manufacturing roadmap. The remaining question is how quickly the company can translate those investments into qualified production capacity and shipments as hyperscalers migrate toward 800G and 1.6T connectivity.

Applied Optoelectronics (AOI) — Company Profile

Leadership — Dr. Thompson Lin — Founder, President & CEO

AI Optics — 400G, 800G and 1.6T optical transceivers; external laser technology and optical components

Recent Milestone — October 2026: Completed $600M ATM program, generating approximately $588M net

Related Converge Digest Coverage — Applied Optoelectronics

SEPTEMBER 30, 2026 · OPTICAL AOI Debuts 25G Tunable DWDM SFP AOI expands its transport-optics portfolio while marking its millionth 1.8 GHz HFC amplifier shipment.

AUGUST 9, 2026 · CORPORATE STRATEGIES AOI’s 800G Ramp Accelerates as U.S. Manufacturing Gains Strategic Weight Q2 analysis examines AOI’s accelerating 800G ramp, capacity constraints and expanding Texas manufacturing footprint.

JULY 14, 2026 · OPTICAL AOI Starts Pearland Factory Buildout for 800G and 1.6T Optics Nearly 400,000 square feet of additional Texas manufacturing capacity targets next-generation data center optics.

APRIL 29, 2026 · CORPORATE STRATEGIES AOI Lands $20.9M Texas Grant to Scale Transceiver Manufacturing Texas Semiconductor Innovation Fund support backs AOI’s domestic high-speed transceiver manufacturing expansion.

MARCH 9, 2026 · OPTICAL AOI Lands $200M Hyperscaler Order for Transceivers AOI’s first major 1.6T volume order established the demand backdrop for its current manufacturing expansion.