$1,000 follows their calls into $997
Daniel Koss@daniel_kossKraken Robotics ($PNG.V / $KRKNF) stands out like crazy when you compare defense stocks by growth and profitability. But what happens when we adjust for valuation? It STILL stands out. You get great growth + profitability at a very attractive valuation.
Daniel Koss@daniel_kossGoldman models Google's compute deal with SpaceX at ~$45B per GW per year. Goldman's own math says hyperscalers only need ~$12B per GW per year to earn a 15% return on capacity they build themselves. Paying ~4x that only makes sense if building it yourself fast enough isn't an
Shay Boloor@StockSavvyShayGoldman estimates short-term AI compute is fetching $40B to $50B per GW annually across $SPCX, $NBIS and $CRWV nearly double long term rates and ~20x what a powered data center shell earns. What makes this really interesting is Nebius captures both spot and contracted pricing
Daniel Koss@daniel_kossJensen is building his answer to TPUs and Trainium. The closed labs keep moving more of their compute onto custom chips. Open-weight models run on Nvidia, on any cloud. Now he's in talks to buy Reflection, which already has a $1B+ compute deal with Nebius. The more AI runs on
zerohedge@zerohedge*NVIDIA IN TALKS TO ACQUIRE REFLECTION AI: FT Jensen going all-in open models.
Daniel Koss@daniel_kossI always like $NBIS fundamentally, spiritually and sexually. When I trim some shares, it always just means I don't like it technically, never anything else.
Daniel Koss@daniel_kossArtificial Analysis just measured AI agent workloads. Tokens per second, per megawatt: NVIDIA GB300: ~261M B300: ~220M AMD MI355X: ~128M 2x the output on the best stack. Microsoft sized its new Maia 300 chip the same way, in tokens per megawatt. Every chip generation gets
Artificial Analysis@ArtificialAnlysWe’ve added DeepSeek V4.1 Flash to AA-AgentPerf, with results for NVIDIA and AMD hardware. At 100 tokens/s per user, NVIDIA GB300 is in first place with throughput up to 261M tokens/s per MW. AA-AgentPerf is our agentic inference performance benchmark. It replays real agentic
