I’m curious when a space elevator would become practical if such a consistent or increasing demand for tonnes to orbit exists

If Darwin was alive I wonder how theory of natural selection will evolve to include this life hack?

Predictions like these can benefit from more clarity if they remember that US is not the only country that uses “dollar” in their currency name. 🤷

Clustering NVIDIA DGX Spark + M3 Ultra Mac Studio for 4x faster LLM inference. DGX Spark: 128GB @ 273GB/s, 100 TFLOPS (fp16), $3,999 M3 Ultra: 256GB @ 819GB/s, 26 TFLOPS (fp16), $5,599 The DGX Spark has 3x less memory bandwidth than the M3 Ultra but 4x more FLOPS. By running compute-bound prefill on the DGX Spark, memory-bound decode on the M3 Ultra, and streaming the KV cache over 10GbE, we are able to get the best of both hardware with massive speedups. Short explanation in this thread & link to full blog post below.

Bun v1.3 is our biggest release yet

AI Bubble, Bubble, Toil and Trouble? It's all the rage for the media today to proclaim an AI bubble. In particular, Bloomberg with this article (linked in the comments below). Here’s a detailed analysis of where and why Bloomberg’s “OpenAI, Nvidia Fuel $1 Trillion AI Market With Web of Circular Deals” is misleading, wrong, or intellectually sloppy. 1. Misleading framing: “Circular” ≠ “Self-dealing” Bloomberg frames the Nvidia–OpenAI–AMD–Oracle ecosystem as “circular” - implying a kind of self-reinforcing illusion of demand. But in reality, these are vertical supply chain linkages, not circular money laundering loops. Nvidia sells GPUs → OpenAI buys them → OpenAI sells API services → revenue funds more compute. Oracle or CoreWeave buy chips → rent GPU capacity → serve paying enterprise customers. Those customers (Microsoft, JPMorgan, Tesla, etc.) generate actual cash flow. There’s no evidence of fake demand or round-tripping. The capital flows are investment → infrastructure → service → revenue, not investment → self-purchase → inflated asset (which would be circular). Bloomberg collapses this distinction entirely. 2. Fabricated or exaggerated deal sizes (“$100B,” “$300B,” “$1T”) The headline figures are fantasy-level aggregates with no substantiated documentation. Nvidia’s supposed “$100 billion investment in OpenAI” has never been confirmed by SEC filings, Nvidia statements, or OpenAI board disclosures. Bloomberg likely conflated total data-center buildout cost projections (which might reach $100B) with Nvidia’s capital investment. Nvidia does not make direct $100B equity investments in anyone. Its total free cash flow for FY2025 is roughly $65B. The “$300 billion Oracle deal” is similarly a gross contract value across years of capacity, not cash spent. It’s analogous to AWS–OpenAI reserved-instance agreements, not a single transaction. “$1 trillion AI boom” mixes cumulative projected CapEx and market capitalization changes and total contract values - triple counting the same dollars at multiple levels of the stack. Bloomberg’s arithmetic exaggerates by an order of magnitude. 3. Confusing CapEx, OpEx, and equity investment The story continually blurs capital expenditure (data-center build) with equity investment and revenue contracts. Example: “OpenAI struck a $300B deal with Oracle… Oracle, in turn, is spending billions on Nvidia chips, sending money back to Nvidia.” That’s not circular finance - it’s normal industrial layering. Each actor’s spend becomes another’s revenue, just like: Boeing → engine supplier → parts supplier → steel producer. Calling this a “web of circular deals” is like calling the auto industry a Ponzi scheme because GM buys from Bosch who buys from ArcelorMittal. 4. Ignores genuine downstream demand AI infrastructure is not speculative inventory. GPU clusters are immediately rented to paying customers (enterprises using Copilot, Midjourney, Databricks, Tesla FSD labeling, etc.). This differs fundamentally from dot-com “click fraud” or unsold banner ads. By ignoring that end-user demand for AI inference and training is real and monetizable, Bloomberg presents growth as hollow. Cloud GPU utilization rates are above 90% across major providers - hard evidence against the “bubble built on circular deals” thesis. 5. Historical false equivalence (dot-com bubble analogy) “In the late 1990s, circular deals were often centered on advertising and cross-selling...” That analogy fails because: Dot-com firms inflated revenues through barter (A buys ads from B who buys from A). Nvidia/OpenAI deals involve physical assets, depreciation, and cash payments tracked on audited balance sheets. AI hardware has salvage value; data centers are tangible productive assets, not vapor. So the analogy is rhetorically catchy, but economically nonsensical. 6. Ignores profitability asymmetry Bloomberg says: “Never before has so much money been spent… on a technology that remains unproven as an avenue for profit.” False: Nvidia, TSMC, and Microsoft are already generating tens of billions in profit from AI. Nvidia’s gross margin >70%; its $4.5T market cap is underpinned by actual $120B+ annualized revenue. Cloud providers’ AI services (Azure OpenAI, Amazon Bedrock, Google Vertex) are profitable at the infrastructure layer. The software-startup layer (OpenAI, Anthropic, xAI) is cash-burning, but that’s standard frontier CapEx - not evidence of system-wide unprofitability. 7. False causality: “Interconnected = inflated” The piece implies that because these companies invest in each other, the market is “artificially inflated.” But cross-investment is standard in high-tech ecosystems: TSMC and Apple co-invest in fabs. Samsung supplies Apple OLED panels while competing in phones. Microsoft and OpenAI’s reciprocal investments are structured as revenue-sharing, not circular financing. Such mutual dependencies actually stabilize supply chains and accelerate deployment - the opposite of speculative froth. 8. Omitting technological fundamentals The article never mentions: The AI compute demand curve, doubling every ~6 months. Model size growth, inference latency requirements, or energy scaling constraints. How the “Stargate” buildout corresponds to actual model roadmaps (GPT-5, 6, 7, multimodal training). Without this context, Bloomberg mistakes infrastructure scaling for financial gimmickry. 9. Misunderstanding SPV and structured-finance mechanics “xAI’s $20 billion round… structured via a special purpose vehicle… to buy Nvidia processors.” That’s normal project financing - an SPV buys equipment and leases it back (like aircraft leasing or power-plant project finance). It’s not evidence of a loop; it’s asset-backed lending. Every hyperscaler uses SPVs for depreciation and risk isolation. 10. Cherry-picked pessimistic quotes, omitting balancing facts Uses Morningstar and Harvard academics to raise “bubble” flags. Ignores that every major investment bank (Goldman, JPM, Morgan Stanley) forecasts continued double-digit AI infrastructure growth through 2030. Ignores explicit denials by Nvidia and AMD that investments are conditional on chip purchases (included, but buried). This selection bias amplifies fear without balanced context. 11. Basic math errors about Nvidia’s capacity Claim: “Nvidia has crested the AI wave… with a $4.5 trillion market cap… investing $100B in OpenAI…” If Nvidia were investing $100B cash in OpenAI, it would consume nearly two years of total free cash flow and trigger massive SEC disclosures - none exist. Bloomberg’s own “PitchBook data” later cites $2B-level equity stakes - a 50× discrepancy. That’s not nuance; that’s factually incorrect reporting. 12. Ignores national-strategic context Bloomberg frames the U.S. government’s laissez-faire stance as negligence. But: The CHIPS Act, DoD compute reserves, and DOE grid partnerships are deliberate national-scale moves to ensure U.S. AI dominance. “Circular” private investments are part of a coordinated industrial policy (similar to defense procurement cycles), not random bubble behavior. 13. Misrepresents OpenAI’s burn and funding “OpenAI… burning through cash and doesn’t expect to be cash-flow positive until near the end of the decade.” That’s a paraphrase of older comments (2023–2024). Recent reports show OpenAI profitable on a gross-margin basis from API and enterprise deals; the losses are from expansion CapEx. Bloomberg conflates operating loss (due to investment) with negative unit economics (which are not true). 14. Equating hype with fraud The rhetorical climax - “Altman could crash the global economy” - is sensationalist. Even a total OpenAI collapse would shave <0.1% off global GDP. AI CapEx is <2% of U.S. corporate investment - large, but not systemic. There’s no leverage or contagion mechanism akin to subprime CDOs or dot-com debt. So the “systemic bubble risk” narrative is economically false. Bottom line Why right an accurate article when a sensational one gets more attention? The truth is self-evident.

As foretold my Tesla Model Y 100% range is now 420km !

Elon Musk on Why Hands are the Key to Humanoid Robots Tesla has launched an official account on Douyin, China's version of TikTok, and has again shared Elon Musk's emphasis on the importance of robot hands. According to Musk, a humanoid robot must first have a pair of "good hands," which he describes as precise machines and extremely versatile tools. He believes the evolution of human hands is close to perfect. With most muscles located in our forearms, the hands are in some ways like a "marionette," capable of performing a vast number of complex and delicate movements. Depending on the calculation, each hand has approximately 27 or 28 degrees of freedom. If you want a robot to truly be able to do all the things a human can do, you must first solve the "hand" problem.

now you can run real-time object detection on multiple streams with 10 lines of code link: github.com/roboflow/inferenc… ↓ code snippet

I can not stress enough how tough it is to build a web app that works well across browsers, mobile devices, and input methods while staying accessible and performant. Hug your frontend developer.

has anyone talked about this part of AI-generated code yet?

After consuming multiple articles/videos/podcasts what I gather is nobody have a full understanding as to how exactly LLMs are capable of ALL the magic they do. Humanity however have figured out a way to lobotomise/fine-tune it to adhere to their respective biases. Its fascinating to see advances just with latent space activation and various prompting strategies.

I would 1. Gather context from author with 1.1 Extra documentation 1.2 Video demos 1.3 Code comments 2. Check out code in local and test 2.1 Assert if use cases are meeting expectations 2.2 Check regressions to other areas 3. Leave helpful constructive review comments

Javascript being able to outperform some is indeed a reason to celebrate ♥️

Mixtral is simply mind blowing for anything that I use GPT 3.5 for !!!

Amazing open feedback loop … in just under 9 hours

Biome 1.4 is out and officially claims the $22.5k Prettier bounty! Read our announcement for more details. biomejs.dev/blog/biome-wins-…

We don't see reality, we see our reality.

Grab it while the offer lasts !

Both the video as well as voice seems to be AI generated to me