As capability gaps between frontier models narrow, inference cost and per-token pricing—not benchmark scores—are shaping the pricing strategies of AI labs.
Below-threshold error correction has reached peer-reviewed hardware demonstrations, but qubit coherence still sets the resource budget for fault tolerance.
As capability gaps between frontier models narrow, inference cost and per-token pricing—not benchmark scores—are shaping the pricing strategies of AI labs.
DeepSeek and Qwen are pushing inference costs down to levels that are forcing Western labs to reposition their pricing, even though a reasoning gap remains in some categories.
🇨🇳 Shenzhen
ON THE GROUND
DeepSeek and Qwen are pushing inference costs down to levels that are forcing Western labs to reposition their pricing, even though a reasoning gap remains in some categories.
| Company | Round | Amount | Valuation |
|---|---|---|---|
| Xendit🇮🇩 Indonesia · Fintech | Series D | 220M USD | 2,4B |
| Saigon Silicon🇻🇳 Vietnam · Chip design | Series B | 48M USD | 310M |
| Krutrim🇮🇳 India · AI | Series C | 1.200M USD | 7,1B |
| Moonshot AI🇨🇳 China · LLM | Series B+ | 600M USD | 4,8B |
| Tomoru🇯🇵 Japan · Robotics | Seed | 14M USD | non-disc. |
| # | Model | Score | $/1M tok |
|---|---|---|---|
| 1 | DeepSeek V4🇨🇳 DeepSeek · 24/05 | 94,3 | 0,27 |
| 2 | Claude Opus 5🇺🇸 Anthropic · 12/05 | 92,8 | 15,00 |
| 3 | GPT-5🇺🇸 OpenAI · 03/04 | 91,2 | 10,00 |
| 4 | Gemini 3 Ultra🇺🇸 Google · 28/03 | 89,4 | 7,00 |
| 5 | Qwen 3 Max🇨🇳 Alibaba · 15/05 | 88,1 | 0,42 |
From Jakarta to Ho Chi Minh City, governments are racing to set up “AI sandboxes” with looser testing conditions than the EU or the US. Behind that wave sits a thorny question: who will control the data that 680 million Southeast Asians are about to feed into these models?
DTW interviewed 31 sources — 12 officials, 9 investors, 6 founders, 4 researchers — to redraw the map of a negotiation that is shaping the region’s next decade.
DTW DEEP DIVE №24
Southeast Asia · AI capital flow
Now in its fifth year, the program honors the 35 most impactful tech founders under 35 across Asia. An independent jury of 14 members from 9 countries; methodology is public.
⏱ 35 days remaining · 412 nominations received
Below-threshold error correction has reached peer-reviewed hardware demonstrations, but qubit coherence still sets the resource budget for fault tolerance.
DeepSeek and Qwen are pushing inference costs down to levels that are forcing Western labs to reposition their pricing, even though a reasoning gap remains in some categories.
A critical CVE patch is the easy part. Whether you can upgrade without breaking production depends on the migration path the fix ships in.
9,1/10
8,1/10
6,8/10
4,2/10
