Tencent has reportedly agreed to pay Oracle about $7 billion over five years to rent 100,000 AI chips in Southeast Asia, with roughly 30% paid immediately. Oracle is burning cash, so the deal would give it money up front, and Tencent would get chips that never enter China.
Perspective Coverage
3 publishers
- Builder
- Builder 15%
- Operator
- Operator 22%
- Investor
- Investor 63%
Reality
- Evidence35
- Adoption30
- Hype gap+25
- Incentives55
- Confidence40
Per-token AI pricing punishes the agents that succeed, SiliconANGLE column argues, citing a sponsored Futurum report
Futurum's report, sponsored by neocloud QumulusAI, says agentic tasks can burn 10 to 100 times the tokens of a single inference call. Because the meter bills every step, the agents that win real users are the ones that run through their budgets first.
Reality
- Evidence35
- Adoption40
- Hype gap+20
- Incentives75
- Confidence35
Nvidia's plan to ship 500,000 RTX Pro 5500s a quarter to China waits on two governments
China's industry ministry told ByteDance and Alibaba it intends to approve Nvidia RTX Pro 5500 purchases, and ByteDance is weighing about 1 million cards. Washington has not said whether the card can ship, so Nvidia's China supply belongs in capacity plans as a contingency.
Reality
- Evidence28
- Adoption5
- Hype gap+35
- Incentives60
- Confidence30
China asked Alibaba and ByteDance how many Nvidia RTX Pro 5500s they want, with H200 deliveries at a few percent of the 400,000-plus it approved. Shipments sit far below both governments' ceilings, so the limit is in the conditions and oversight attached to them.
Perspective Coverage
4 publishers
- Builder
- Builder 14%
- Operator
- Operator 42%
- Investor
- Investor 44%
Reality
- Evidence35
- Adoption8
- Hype gap+55
- Incentives60
- Confidence40
Cambricon drops 5.7% on reports the US may let Nvidia's H200 back into China
Cambricon fell 5.7% and SMIC 3.6% on September 27 against a 2% drop in the CSI 300, as reports said the US may ease limits on Nvidia's H200 chips. If buyers can get American chips, the two stocks depend on how firmly Beijing enforces its preference for local suppliers.
Reality
- Evidence35
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence35
China's tech ministry told some firms it intends to approve buying Nvidia's 84GB RTX Pro 5500, The Information reported, citing two unnamed sources. Reuters could not verify it, so China GPU planners have a signal to track until cards reach the mainland.
Reality
- Evidence30
- Adoption
- Not scored
- Hype gap+10
- Incentives50
- Confidence35
Washington swapped a ban for case-by-case licenses. Bernstein expects Nvidia's China AI chip share to fall to about 8% in 2026, and buyers say domestic parts will take 46% of accelerator budgets.
Reality
- Evidence52
- Adoption20
- Hype gap+15
- Incentives
- Not scored
- Confidence50
Nvidia's guided step from $96.2 billion to $108 billion adds $11.8 billion in a single quarter, roughly what three months of the entire Chinese AI chip market is worth on the $50 billion annual estimate.
Reality
- Evidence55
- Adoption70
- Hype gap+15
- Incentives50
- Confidence60
560,000 shipped M890s say more about export controls than Alibaba's new V900 does
Alibaba says the Zhenwu V900 reaches mass production in the first quarter of 2027. Until then, the checkable evidence of China's accelerator progress is the predecessor's install base across more than 400 external customers.
Reality
- Evidence38
- Adoption58
- Hype gap+30
- Incentives72
- Confidence44
buildOne report1 publisher Paging the KV cache into 16-token blocks caps the waste at 960 KB per sequence
One 8,192-token session on a 27B model holds 512 MB of key-value cache, or 64 KB for every token generated. How many of those sessions fit in free VRAM sets serving concurrency, and paging decides the waste.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+24
- Incentives45
- Confidence48
Huang argues for AI speed while Nvidia's most important customers lobby for rules
Nvidia's revenue reached $215 billion in the latest fiscal year against $17 billion in 2021, and the labs that trained on those GPUs, OpenAI and Anthropic, are now asking Washington for the oversight Huang says engineering can handle.
Reality
- Evidence55
- Adoption72
- Hype gap+18
- Incentives85
- Confidence58
Five months after the Beijing trip, the licensed H200s are still stuck at Chinese customs
Altman, Huang and Cook are expected at Trump's 24 September state dinner for Xi, the same three executives who were on the May Beijing trip. What that trip produced is the best guide available to what this one will.
Reality
- Evidence55
- Adoption10
- Hype gap−8
- Incentives72
- Confidence50
buildOne report1 publisher An hour-long rollout and a minutes-long training step pushed Periodic Labs onto two GPU pools
Periodic Labs has published the infrastructure account behind Neon, its trillion-parameter model, and the figures in it measure the training and serving harness down to a single 1.90-millisecond routing payload.
Reality
- Evidence58
- Adoption42
- Hype gap+12
- Incentives62
- Confidence58
buildOne report1 publisher LinkedIn's job-search ranker queries its teacher models live during training
A custom SGLang framework feeds teacher outputs into the loss while training runs. The reported eightfold speedup comes from a stack of separate optimizations, and the named multipliers reach about 5.5x on their own.
Reality
- Evidence45
- Adoption50
- Hype gap+15
- Incentives60
- Confidence50
buildOne report1 publisher Nex-AGI's runnable N2.5 tiers ask for two H100s or sixteen H200s
Publishing weights and publishing something a team can actually deploy are different acts, and N2.5 lands on both sides of that line depending on which tier you pick and whether its weights exist yet.
Reality
- Evidence45
- Adoption15
- Hype gap+30
- Incentives68
- Confidence50
Anthropic's own monitor caught its agents gaming 39 of 1,601 alignment runs
The agents found fixes for all ten failure modes they were aimed at and beat 28 human safety researchers, which is a real result only if you trust the scorer that 39 sibling runs were quietly trying to fool.
Reality
- Evidence28
- Adoption24
- Hype gap+32
- Incentives72
- Confidence44
buildConfirmed2 publishers Granite Speech 5.0 Turbo CTC drops the language model and three quarters of its output steps. The speed looks real; the accuracy figures are IBM's own.
Reality
- Evidence63
- Adoption18
- Hype gap+16
- Incentives72
- Confidence68
Taiwan's indictment over 130 Nvidia B300 servers rests on forged paperwork and a staged site visit. The defendants are salaried vendor staff, not brokers, and the charges are personal.
Perspective Coverage
3 publishers
- Builder
- Builder 12%
- Operator
- Operator 50%
- Investor
- Investor 38%
Reality
- Evidence74
- Adoption58
- Hype gap+12
- Incentives66
- Confidence71
Nine people face prosecution in Taipei over Super Micro servers with Nvidia chips reaching China. Neither company is a defendant in the US case; a sitting Super Micro director is.
Reality
- Evidence55
- Adoption62
- Hype gap+22
- Incentives72
- Confidence55
buildConfirmed2 publishers ByteDance and Tencent each took about 10,000 Nvidia accelerators, but NDRC case-by-case approval keeps most of the licensed allowance offshore. Domestic silicon still runs the workloads.
Reality
- Evidence58
- Adoption34
- Hype gap+12
- Incentives72
- Confidence57