JaMarco

@SingularityRes @HotAisle I can think of a way that happened

@HotAisle Last year DP was saying Nvidia was specsmaxxing their GPU Today Jensen Hwang said DP told him that Nvidia was sandbagging their performance I don’t know how DP was able to say one thing to the world and totally different to what he said in private

I hope MC replaces semianalysis.

@danforhan @nw9824 @TermineRadio No because it was a league ownership decision

@nw9824 @TermineRadio And none of the other 28 teams came out and said anything?

Luka basically with more points in the last 7 days than Anthony Davis has had in the year + since the deal. Most senseless trade in nba history somehow makes even less sense today than the day it was made.

@vikramskr @benthompson I think he's saying the opposite of what you're saying

.@benthompson asks Jensen how it feels to be a CPU salesman 🤣 Agentic AI CPUs need the highest core count, not per-core perf as first priority. This is why $AMD Venice is killer: 256 cores, 512 threads per CPU I have a whole catalog of CPU specs on Substack - the AI Server CPU Yellow Pages! open.substack.com/pub/viksnewsle…

@Munkboi69 @AdinUpdate he hates neon?

@AdinUpdate 100% Conor McGregor

Iggy Azalea tells N3on a guy once sent her a d-ck pic after she said she was having a bad day… N3on then asks to SEE it and finds out it’s someone FAMOUS 😭 “Oh my god, that could sell for a lot of money.”

@PepInvestStocks @RealJGBanks How?

@RealJGBanks I prefer $SMR, more upside potential.

BREAKING: $OKLO just made the nuclear trade bigger Oklo is building an isotope reactor in Texas This adds a new layer to the nuclear theme: SMR / Advanced nuclear next-gen reactors $OKLO $SMR $BWXT $LTBR Fuel / enrichment supply bottleneck $CCJ $UEC $LEU Power / grid AI electricity demand $CEG $VST $PWR Medical isotopes new growth market $BWXT $LEU This proves the nuclear trade isn’t over It’s just starting.

@ClockworkTerps @Electron_Cowboy @CrusoeAI Geez

@Electron_Cowboy @CrusoeAI I can’t wait for the AI slop garbage bubble to pop and for you to lose everything

New video released for@Nvidia GTC showcases @crusoeai facilities and operations around the world: - Geothermal-powered Crusoe Cloud data centers in Iceland - Electrical component factories in Colorado, Oklahoma and Louisiana - GW+ AI training campuses powered by gas and wind power plants in Texas - Modular Spark data centers powered by solar and used EV batteries in Nevada youtu.be/0i18rms-fPM?si…

@benitoz @firstadopter Wasn't cpx for prefill?

@firstadopter Once the LPU came out, there was no need for CPX. Custom ASICs are DOA.

Nvidia CPX is NOT coming out in 2026: Nvidia executive.

@comic @Pledditor @jackmallers awful look? Its complete collapse

@Pledditor @jackmallers That’s clearly not happening. Tether and SoftBank resigning is an awful look.

@comic @jackmallers lord have mercy

@EnerTuition @DylanOnChips @DreadyBear aws bought an expert chip company though

@JaMarc0 @DylanOnChips @DreadyBear AWS guys thinking that they can compete with merchant semi guys in a fast-moving space is a fallacy. All else being equal, companies who stick to their core competency will do better in the long term.

Jensen: "There are no CPUs in the world that are two times the performance of anything else, besides Vera." He doesn't see Nvidia's CPUs playing in the traditional server CPU market: "That's not the problem we're trying to solve."

@EnerTuition @DylanOnChips @DreadyBear who is on the right path?

@DylanOnChips @DreadyBear By the way, Amazon is on the wrong path with Graviton in my view (I have written about it). Graviton is not a competitive chip. Amazon's customers are getting screwed and one day they may find out.

@thewhiteboxAI @benitoz @vikramskr huh?

What Vikram is too polite to tell you explicitly is that you're wrong by assuming memory is the bottleneck in highly-agentic workloads. In those instances, what really matters is core CPU count, and in that, AMD is ahead. Where memory matters the most is deep-reasoning agentic workloads, which are by definition "less agentic" than the others.

Not sure I buy the “Vera is best for Agentic AI narrative” by Nvidia - but yes, the cpu flood is coming.

@HotAisle amd pretty much has to buy cerebras at some point

@beefcubee @Mar364503 they only had 3 months to integrate the grok rack. Feynman grok rack will almost certainly use Nvidia custom ARM

@Mar364503 it just bizarre to not use vera cpu on their flagship inference chip

As I said, Jensen did not mention what CPU used for Groq tray because it is likely $INTC CPU. As AI industry moves from training to inferencing, Groq may have a good sale

@Alex_Intel_ @labubu_trader What about arm?

@labubu_trader Long AMD for CPU is dumb Rated limited by TSMC. Intel can expand most supply in next 18 months

I'm accumulating AMD leap calls but still small position. I think the market underestimates AMD's CPU advantage in the AI agent world and its catch-up potential in model inference. ROCm is a joke now compared to CUDA.But with Claude Code's help, I think the gap will close much faster than people can imagine

@davidmarcus @spark @tether @paoloardoino Huge partnership

.@spark in WDK is a big milestone for Bitcoin. Thrilled about our partnership with @tether to enable hundreds of millions of wallets with fast, cheap Bitcoin payments. 🤝 @paoloardoino & team

@Z3TURIN @perry_lin1 Good question

@perry_lin1 Hi Perry, do you still have exposure on DEFT ? Looks like it has bottomed and the next crypto leg could reverse the trend Thanks in advance

$PUMP $15 incoming

@cwa93393 @rohanpaul_ai 🤣

@rohanpaul_ai what would digital employees do with the crypto? That makes no sense. They don’t have any personal expenses, there’s no incentive to have money. People in crypto are way over their heads with this “future of money” thing.

Coinbase CEO: AI Agents write over 50% of code and resolve 60% of support tickets. To scale autonomy, they provide agents with stablecoin wallets for machine-to-machine payments, and really treat them like digital employees.

@bidask123 @BTCoptioneer 🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣

@BTCoptioneer ASST 1 for 20 reverse split. MSTR no 1 for 20 reverse split.

The only problem with $ASST is the amplification ratio 44% is too high and they need Bitcoin to rally before issuing more SATA

@BillQueens @andrewdfeldman @cerebras but thats just a financing issue

@JaMarc0 @andrewdfeldman @cerebras They don’t even have enough compute for you to sign up to be a customer right now

NVIDIA's biggest GTC announcement was a $20 billion bet on the same problem we solved 6 years ago. Their next-gen inference chip - not available yet - has 140x less memory bandwidth than @cerebras. To run a single 2 trillion parameter model, you need 2,000+ Groq chips. On Cerebras, that's just over 20 wafers. Even paired with GPUs, Groq maxes out at ~1,000 tokens per second. We run at thousands of tokens per second today. And every day. In production now. Why? When you connect 2,000 chips together, every interconnect has latency. Every cable has overhead. It doesn't matter what your memory bandwidth is on paper if you're bottlenecked by the wiring between thousands of tiny chips. We solved this with wafer scale. One integrated system. Little interconnect tax. Jensen told the world that fast inference is where the value is. He’s right - it’s why the world’s leading AI companies and hyperscalers are choosing Cerebras.