Pick up an NVIDIA B200 and trace where the roughly 6,400 manufacturing cost actually goes.
NVIDIA's Blackwell B200 debuted at 0.11 per million tokens on SemiAnalysis's InferenceMAX benchmarks.
NVIDIA quietly shipped one of the most interesting open model architectures I've seen this year, and most of the coverage buried the lede.
AMD just crossed a threshold that matters more than any spec sheet: one million tokens per second from a single cluster, verified by MLPerf.
Most open models fight for the same throne: highest MMLU score, best SWE-Bench pass rate, flashiest reasoning demo.
Every robotics lab has the same complaint: the model architecture works, the sim environment works, but there's never enough training data.
A chip vendor hiking prices 67% overnight would normally send customers scrambling.
Most open model launches follow a predictable script: bigger parameters, higher benchmark scores, a claim about beating GPT-something on MMLU.
Amazon will spend 200 billion on capital expenditure this year. Alphabet somewhere around 175 billion.
Google buried the announcement inside a Q1 earnings call that had plenty of other headline-worthy numbers — 109.
Every multimodal agent demo I've seen follows the same script: chain a vision model to a language model, bolt on Whisper for audio, pray the latency...
Scott Guthrie called Maia 200 "30 percent cheaper than any other AI silicon on the market" when he unveiled it in January.
NVIDIA's GR00T N1 humanoid robot trained on 780,000 synthetic episodes — 6,500 hours of practice generated in just 11 hours.
NVIDIA doesn't do charity — so when Jensen Huang wrote a 5 billion check for 214 million Intel shares at 23.
Epoch AI published a manufacturing teardown of NVIDIA's B200 last month.
Jensen Huang called it a "horrible outcome for America.
Scroll through the investor list on SiFive's freshly closed 400 million round and you hit a name that shouldn't be there: NVIDIA.
Google just told you its biggest cost problem isn't training. It's serving.
Quantum computing has a plumbing problem.
Jensen Huang does not spend $20 billion on chips he could build himself. He spends it on chips he did not want to admit he needed.