"NVIDIA DSX is full stack AI factory reference architecture, co-innovated with the world’s leading infrastructure companies." IREN sites in Jensen Huang's #NVIDIAGTC Berlin keynote. Watch the full presentation: youtube.com/watch?v=PvFjbPYE…

WHY $CIFR COULD BE THE BIGGEST BARGAIN IN AI INFRASTRUCTURE RIGHT NOW I think Cipher could be emerging as one of the most interesting opportunities in the entire AI infrastructure selloff because the market has become so focused on leverage, dilution and financing risks that it's overlooking how dramatically the underlying business is transforming. Cipher has fallen nearly 70% from its previous highs, yet the company is no longer primarily a Bitcoin miner and is increasingly becoming a large-scale AI data center developer and infrastructure landlord with approximately 5.3 GW of potential capacity across 11 projects. That's an important distinction because unlike neoclouds that borrow billions of dollars to purchase GPUs and rent compute capacity, Cipher is developing the physical infrastructure that hyperscalers need while its tenants provide the expensive computing equipment themselves. This means Cipher isn't directly exposed to GPU depreciation, rental pricing or the constant replacement cycles that can make the economics of compute providers so difficult to underwrite. Instead, its opportunity is increasingly tied to securing scarce power capacity, constructing data centers and generating recurring rental income through long-term agreements with some of the largest technology companies in the world. I think investors are making the mistake of treating every leveraged AI infrastructure company as though it has the same business model, when there's an enormous difference between financing rapidly depreciating GPUs and developing infrastructure that could generate contracted cash flows for the next 15 to 20 years. What makes the opportunity particularly compelling is that we're finally beginning to see Cipher transition from announcing enormous development pipelines into actually delivering infrastructure and collecting rent since Black Pearl has already begun generating rental income after its initial capacity was delivered ahead of schedule, while Stingray has secured approximately $810 million in project financing at a 6% coupon, covering nearly 98% of expected project costs and backed by a 15-year lease with annual rental escalators of approximately 3%. That financing was reportedly oversubscribed by roughly eight times, which I think demonstrates that lenders remain willing to finance high-quality AI infrastructure projects when the underlying contractual economics are attractive. Then you have Barber Lake, where the lease extension increased the contracted term from 10 to 20 years and expanded nominal contracted revenue from approximately $3.8 billion to more than $9 billion. Across its initial leased projects, Cipher has outlined hundreds of millions of dollars in expected annualized net operating income, while its broader development pipeline provides the possibility of signing additional long-term agreements as hyperscalers continue competing for available power and data center capacity. I think this is where investors are missing the bigger picture because we're talking about infrastructure with potentially decades of revenue visibility, and the market is increasingly valuing it as though the underlying AI investment cycle could collapse before these facilities even begin generating meaningful cash flow. Now, I don't want to pretend the recent selloff is entirely irrational because the biggest challenge for Cipher is that securing enormous contracted revenue doesn't automatically translate into attractive returns for shareholders. Barber Lake is probably the clearest example because the projected construction budget increased approximately 36% to 38%, pushing expected project costs toward $3 billion while reducing the estimated first-year cash NOI yield from approximately 13.5% to 14% down to 10.5%. That's a meaningful deterioration in capital efficiency, even though the contractual revenue opportunity became substantially larger. The company also recently announced approximately $850 million in zero-coupon convertible financing alongside an exchange of existing debt for roughly 40.7 million shares, which reinforces that dilution and capital structure complexity remain legitimate concerns. However, I think there's an important distinction between financing that becomes expensive because a business is struggling to find customers and financing used to complete infrastructure that already has long-term contractual commitments. Raising capital at a 0% coupon in an environment where Treasury yields are above 5% is attractive financing on its face, even after accounting for future conversion risk, while project-level debt supported by tenant contracts can offer more favorable economics than unsecured corporate borrowing. The real question is whether Cipher can consistently deliver projects at costs that generate attractive returns on the equity invested, and that's why I'm paying considerably more attention to construction spending, financing terms and cash yields than simply celebrating every additional billion dollars in announced backlog. Ultimately, I think the potential deal-of-a-lifetime opportunity in Cipher comes from the disconnect between what investors are currently willing to pay for the business and what its infrastructure could eventually generate if management successfully executes. At approximately $9 per share, Cipher's pro forma equity valuation is around $4.1 billion, with an indicative enterprise value of approximately $5.5 billion against roughly $686 million in projected 2027 contracted net operating income, implying a multiple of approximately 8x before accounting for corporate expenses, financing costs and additional development requirements. That certainly doesn't make the stock automatically cheap because project-level NOI isn't the same thing as free cash flow available to shareholders, but I think it illustrates how aggressively expectations have compressed relative to the company's contracted opportunities. The bigger upside would come from successfully delivering Black Pearl, Barber Lake and Stingray while converting additional portions of its multigigawatt development pipeline into signed leases without continually diluting shareholders. If management can demonstrate that construction costs are stabilizing, rental income is accelerating and additional projects can be financed on attractive terms, I think investors could begin valuing Cipher as a growing contracted infrastructure platform rather than a speculative former Bitcoin miner. However, if cost overruns continue or financing requirements increase faster than the cash flows available to shareholders, the stock could remain under pressure regardless of how strong AI demand becomes. That's why I think the next several quarters will be incredibly important because the opportunity is no longer simply about proving that hyperscalers want the capacity, but demonstrating that Cipher can deliver it economically. My thesis is that the market may be pricing in an enormous amount of execution and financing risk at exactly the moment the company is beginning to turn its contracted infrastructure into a cash-generating business, and if that transition succeeds, I think the potential revaluation could be substantial.

$SPCX: "It is not impossible that we will have a six-month payback period for compute next year."

🚨 JUST IN: Nick Shirley just dropped this line in the Oval Office flanked by President Trump, and it's PURE GOLD "I want people to think about what is happening right now...me as a 24-year-old, I went out and exposed the fraud. I just was concerned about the well-being of our tax dollars. And right now, I'm standing next to the President of the United States." 🔥 "There's never been an incentive for anyone to ever expose fraud [until now]...when you expose fraud, you help save America. People understand that these fraudsters take from the poor needy who actually need the services that we all work hard to pay for, and we take that, and they then buy Mercedes!" @nickshirleyy WELL-DESERVED FOR NICK. A year ago, legacy media mocked him as a right-wing propagandist Today he walked into the Oval Office with the first check for exposing fraud that the media IGNORED!! Camera, microphone and passion for saving America. Let's freaking go.

Mirantis is a founding member of the Open Secure AI Alliance, now hosted by the Linux Foundation (@linuxfoundation). An AI system runs on a full stack of model, harness, runtime, and infrastructure. It is only as secure and as open as the layers underneath. We are advancing that work from the infrastructure upward, grounded in what we know best. That means turning hard-won operational experience into open artifacts that others can deploy, test, and trust, instead of keeping it in private runbooks. For the operators who run this infrastructure, you can reproduce a deployment in an environment you control and inspect the evidence behind every claim. See what open AI infrastructure really means. buff.ly/AfSo2tD #Mirantis #k0rdent #OpenSource

Why Working, Episodic, Semantic, and Procedural Memory Matter — and Why AI Memory Is Becoming an Infrastructure Investment Theme .

Holy Moly. Meta has now revealed new details about Prometheus, its gigawatt-scale AI supercluster in Ohio. During synchronized training, Prometheus experiences 30–40% power swings, shifting tens of megawatts in seconds. Meta developed a system called Dynamo Dimmer to throttle GPU power, supported by batteries and capacitors. The first phase has already trained Muse Wonder. Prometheus now connects hundreds of thousands of Nvidia GPUs across multiple buildings, with plans to scale to roughly 1 million and add Nvidia's Vera Rubin hardware. At its peak in mid-2025, Prometheus suffered nearly 1,000 training interruptions per day per 8,000 GPUs. Meta says engineering fixes with Nvidia have since reduced interruptions by 99%. To keep 32,768 GPUs actively training, Meta provisions 37,656 GPUs, including spare capacity for hardware failures.

🚨HOLY. Business Insider obtained internal Amazon emails projecting $5 BILLION in Nvidia GB300 racks deliveries in August, then roughly $3 BILLION EVERY MONTH. Some AWS data centers have room for the racks but NOT enough power to turn them on. Equipment is being staged in warehouses, while Amazon takes ownership earlier, tying up cash before deployment. At an internal September staff meeting, AWS CFO John Felton joked to infrastructure chief Prasad Kalyanaraman: "You've taken us free cash flow negative as a company, so congratulations." AWS's infrastructure chief expects Amazon's free cash flow to remain negative for 'the next couple of years or so.'

Heading to #NVIDIAGTC Berlin? See IREN AI Cloud in action with a demo led by IREN Chief Product Officer, Kambiz Aghili, and learn how IREN brings together NVIDIA GB300 NVL72 infrastructure and k0rdent AI orchestration to support production inference. Join us at the IREN booth 7013 and explore our vertically integrated AI Cloud platform, delivering data centers, compute and software.