Observed Signal · Sep 8, 2026 · Technical Release · Source: Trending Topics (DACH/CEE Innovation & Tech) · Impact: 4/5 · Sentiment: Positive
Magic AI Claims Frontier-Level Pretraining for Under $1M
Magic, an AI startup co-founded by Austrians Eric Steinberger and Sebastian De Ro, claims a major breakthrough in pretraining efficiency. Its new recipe reportedly matches DeepSeek V4 Pro's base model quality using 50x less compute, costing about $500,000 on Nvidia GB200 systems, and is over ten times more compute-efficient than leading open-weight models. Scaling to roughly $4 million, Magic says it outperforms all public base models on perplexity evaluations, comparing against DeepSeek V4 Pro, Kimi K2, and Nvidia's Nemotron 3 Ultra, while excluding closed models from Anthropic, Google, and OpenAI. The company has raised over $460 million, with a $320 million round valuing it at $1.5 billion, and partners with Google Cloud for tens of thousands of GB200 chips. Magic has not yet released a model, and all claims are self-reported.
Magic's claim of frontier-level pretraining at a fraction of the cost could disrupt the AI model training landscape, making it relevant for AI infrastructure and potentially affecting the broader tech ecosystem.
Key Takeaways & Evidence Grounding
- Magic claims its pretraining recipe matches DeepSeek V4 Pro's quality with 50x less compute, costing ~$0.5M on GB200, and is 10x more efficient than leading open models.
- Scaling the recipe to ~$4M, Magic says it beats all public base models on perplexity evaluations.
- Magic has raised over $460M, with a $320M round valuing it at $1.5B, and partners with Google Cloud for GB200 compute.
- Magic has not yet released a model, and its claims have not been independently verified.
- Magic's model achieves a 72% pass rate on math problems in a short RL run, compared to GPT-6 Astra at 100% and Kimi K3 at 75%.
Connected Companies & Entities
10 Entities mappedAtlassian
Enterprise software for collaboration, workflows and team productivity.
“Atlassian backed a $320 million round for Magic....”
Magic
Frontier code-model developer for autonomous software engineering and research.
“Magic is an AI startup founded by Eric Steinberger and Sebastian De Ro, claiming frontier-level pretraining efficiency....”
Fireworks AI
B2B platform for model training, routing and inference.
“Magic brought in the inference provider Fireworks for an independent check of baseline figures....”
Sequoia Capital
Venture capital firm investing in technology companies across stages.
“Sequoia backed a $320 million round for Magic....”
Alphabet
Digital platform conglomerate centred on advertising, media and software.
“Alphabet participated in a $23 million funding round for Magic....”
CapitalG
Alphabet growth fund investing in transformational technology companies.
“CapitalG backed a $320 million round for Magic....”
NVIDIA
Accelerated computing company spanning AI software, cloud and gaming.
“Magic's compute costs refer to Nvidia's GB200 systems, and Nvidia's Nemotron 3 Ultra is used as a comparison model....”
DeepSeek
LLM developer offering AI chat and API access.
“Magic compares its pretraining recipe to DeepSeek V4 Pro and V4 Flash....”
Moonshot AI
Chinese AI company delivering Kimi models, agents, enterprise software and APIs.
“Moonshot's Kimi K2 and Kimi K3 are used as comparison models....”
Google Cloud
Enterprise cloud, data and AI services for businesses.
“Magic announced a partnership with Google Cloud to build compute capacity with tens of thousands of GB200 chips....”
Ontology Mapping & Concepts
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.