Marcus Lemonis reveals concern over US investment in chip companies

IBM announces 0.7-nanometer "alien chip," humanity officially enters the angstrom era! If you haven’t heard yet, follow IBM and Rapidus immediately. Just within the past 24 hours, industry giant IBM officially unveiled the world’s first 0.7-nanometer (7-angstrom) chip technology in New York. While TSMC and NVIDIA are still battling it out in the 2-nanometer flatland, IBM has built a 3D skyscraper in the microscopic world. Is this just a PowerPoint spectacle—or a true “alien technology” capable of overturning the next decade’s computing dominance? Today, we break down this supercharged nuclear bomb that ends Moore’s Law—with hard, factual data. To understand this revolutionary leap in computing power, we must strip away the hype of “0.7 nanometers” and get to the core technical breakthrough. 📊 Core Data Breakdown: Just how terrifying is this “alien chip”? According to IBM’s official disclosures and hard data from major outlets like CNET and CBS, this breakthrough is nothing short of brutal elegance: A universe of hundreds of billions on a fingernail: On a chip no larger than a human thumbnail, IBM has packed a staggering 100 billion transistors. What does that mean? Its density is nearly double that of IBM’s own 2-nanometer chip released in 2021. For perspective, the width of a human red blood cell is about 7,000 nanometers—yet the feature size of these transistors is just 0.7 nanometers (7 angstroms). Dual domination in performance and efficiency: Compared to current 2-nanometer processes, this 0.7-nanometer chip delivers up to 50% higher performance—or reduces power consumption by as much as 70% at equivalent performance levels. For AI data centers drowning in energy crises, this is nothing short of divine intervention. 40% shrinkage in SRAM (Static Random-Access Memory): This marks the first meaningful advancement in SRAM density in over a decade. SRAM is ultra-fast memory—the critical “super brain capacity” AI computing desperately needs. AI compute power surges sixfold: IBM’s research team estimates current top-tier AI accelerators deliver around 1,500 TOPS (trillion operations per second). With 7-angstrom technology, that number could skyrocket to a staggering 9,000 TOPS. 🧠 [Dimensional Analogy]: The Core Breakthrough—The “Nanostack” Architecture Why have all the world’s top engineers been trapped at the 2-nanometer physical limit—until now? The secret isn’t about sharpening a pencil finer. It’s about violent dimensional escalation. IBM has pioneered a revolutionary 3D transistor architecture called “Nanostack.” [Simplified explanation]: Previously, even the most advanced chips (like Nanosheet designs) were built like flat pancakes or single-story houses on a 2D plane. With limited chip real estate, cramming more transistors together leads to massive leakage and overheating. IBM’s Nanostack shifts expansion into the Z-axis—the vertical dimension. It’s like building stacked bunk beds and towering skyscrapers directly on the same footprint, folding transistors vertically. While everyone else fights over flat land, bleeding themselves dry, IBM has ascended into 3D space—and delivered a devastating dimensional strike. 🎯 Investment Insight & Industry Shakeup: Should You Buy Now? At this point, many will think: Sell NVIDIA and TSMC immediately—go all-in on IBM. Hold on, boss! This is the future—but not tomorrow. Wall Street analysts are crystal clear: This is a lab-level foundational breakthrough. It will still take approximately five years before it reaches mass production readiness. The shadow warrior behind the scenes: IBM no longer owns its own chip fabrication plants. To bring this technology to market, it is partnering with Japan’s rising foundry star, Rapidus (Advanced Semiconductor Manufacturing Company). Once matured, Rapidus will become the ultimate assassin threatening TSMC’s global foundry dominance. In summary: Over the next 3–5 years, TSMC’s 2-nanometer production lines and NVIDIA’s GPUs remain irreplaceable cash machines for AI computing. But IBM’s announcement yesterday has delivered a death notice to every semiconductor giant: The era of planar scaling under Moore’s Law is dead. The angstrom era of 3D space has officially arrived. Whoever masters vertical stacking production lines like Nanostack within the next five years will claim the iron throne of the next decade.

The entire thesis rests on a revenue number from a private company that has never disclosed one. Banxia either has a leaked Anthropic investor update, a modelled inference nobody can audit, or something a former employee said in a room. Bloomberg does not know which. Banxia declined to say. That does not make the call wrong. It just means the most important contrarian signal in markets right now cannot be verified until it is already too late to act on it.

27/06/26 Investor’s Daily Brief - AI’s Constraint Trade Gets Real Market snapshot: S&P 500: 7,354.02, -0.05% Nasdaq: 25,297.62, -0.24% 10Y Treasury: 4.372% Oil: WTI $69.23, -3.74% Gold: $4,096.30, +1.20% Bitcoin: $60,521, +0.87% The AI trade did not break today, but it got more constrained. The clean read is that investors are no longer just paying for demand. They are paying for control of scarce inputs: model access, memory, power, labor, infrastructure timing, and regulatory trust. Model access becomes policy risk OpenAI is limiting GPT-5.6 access to approved customers, while Anthropic’s Mythos 5 is being reintroduced only through trusted channels. That changes the AI revenue story because frontier models are no longer pure product launches. They are now permissioned releases. The confirming signal is simple: broad commercial availability arrives quickly. If not, investors have to price government approval as part of the AI stack. Memory becomes the margin transfer The DRAM squeeze is the sharpest earnings signal. Micron gets pricing power, while Apple, Microsoft, console makers, and smaller hardware firms absorb the pain or push it to customers. That is not neutral inflation. It is margin moving upstream. The test now is whether price hikes protect margins without damaging demand. AI infrastructure becomes a timing trade Masa Son questioning orbital AI data centers matters because it reframes the capex debate. Musk may be right over decades, but markets pay for deployable capacity now. Near-term winners still look like terrestrial data centers, chips, power, and distribution. Space labor becomes execution risk SpaceX stock cooling does not mean the space economy stopped hiring. The issue is whether aerospace companies can find enough STEM and skilled manufacturing talent to convert backlog into production. Bottom line AI is no longer just a growth story. It is a constraint story. The winners are the firms closest to scarce inputs. The losers are the firms forced to absorb the squeeze.

💻 OpenAI and Tech Giants Accelerate Custom AI Chip Development to Challenge Nvidia 1. Major AI players including OpenAI, Google, and Amazon are intensifying their efforts to design custom silicon optimized specifically for artificial intelligence workloads. OpenAI recently revealed its new "Jalapeno" chips, reportedly built from scratch in just nine months with assistance from Broadcom. These in-house hardware initiatives aim to drastically reduce the industry's heavy reliance on Nvidia while driving down massive infrastructure costs. Bobbi thinks: Vertical integration is becoming the ultimate defensive moat in AI, as controlling the underlying silicon is now just as critical as training the frontier models themselves. Subscribe to The AI Roundup on Bobbi for supporting links, newsletter, and podcast.

I would be quite happy if Anthropic isn't sucking retail savings with a 1.8T valuation IPO launch while the Chinese are delivering the same models for free (ok, you need good hardware though).

Is Apple lobbying the U.S. government for approval to purchase memory from CXMT, China’s largest DRAM manufacturer? Note: CXMT is on the U.S. military blacklist. End of story.

Oracle Is Learning AI Infrastructure Is Not Software Oracle is falling because the market has stopped rewarding the AI backlog and started pricing the cost of building it. On the surface, the numbers do not look weak. Cloud growth is strong. OCI growth is strong. Remaining performance obligations have surged. In a normal market, that would be treated as proof that Oracle has successfully attached itself to the AI infrastructure boom. But investors are looking past the headline. The backlog does not arrive as free cash flow. It arrives first as a construction bill. That is the entire issue. The Old Oracle Was a Better Business Model The old Oracle was a high margin software machine. Databases, enterprise software, sticky customers, recurring maintenance revenue, and strong cash conversion made it one of the best business models in tech. The new Oracle AI story is different. It is data centers, GPUs, leases, power contracts, cooling systems, land, depreciation, debt, and massive upfront capital spending. That is not asset light software. That is industrial infrastructure with a software multiple attached to it. The Free Cash Flow Problem Oracle can still generate a lot of operating cash flow, but the AI buildout is consuming it faster than the legacy business can replenish it. That is why the market is getting nervous. The company is trying to scale into AI cloud infrastructure at the exact moment when every major player is competing for the same scarce inputs. GPUs, memory, networking equipment, power, data center capacity, and financing are all expensive. So the demand may be real, but the cost of serving that demand is enormous. This is why investors are no longer satisfied with a giant backlog. They want to know what that backlog actually earns after capex, depreciation, debt service, and margin pressure. The Customer Concentration Risk Oracle’s AI cloud story is also heavily tied to a small number of very large customers. That is exciting when confidence is high, because one major AI deal can transform the growth narrative. But it becomes risky when investors start asking harder questions. Can those customers keep spending at this pace? Will the contracts convert into high margin revenue? How much infrastructure must Oracle build before it gets paid back? What happens if AI demand slows, funding tightens, or timelines slip? A huge contract can be valuable, but it can also become a huge obligation if the economics are not clean. The Bigger Software Problem Oracle is also caught between two narratives. Traditional enterprise software is no longer getting the same premium because investors are worried AI agents could eventually pressure parts of the old software stack. At the same time, AI infrastructure requires huge capital spending before the profits are proven. So Oracle is being squeezed from both sides. The legacy software business is being valued more cautiously. The AI infrastructure business is being judged more harshly. That is a difficult place to be. What Has to Happen Next Oracle does not stabilize sustainably until investors believe three things. Capex has a visible ceiling. AI contracts convert into high margin revenue. The company can fund the buildout without permanently damaging free cash flow or diluting shareholders too heavily. Until then, rallies are likely to be sold. My Take The demand is real. The backlog is real. The strategic value is real. But the market is no longer paying for AI dreams in the abstract. It is asking who earns money, who burns cash, who controls scarce capacity, and who is simply writing checks. Oracle is falling because it moved from being valued like a software compounder to being judged like a leveraged AI infrastructure buildout. That is a much harsher market.

Apple seeks US approval to buy memory chips from CXMT 🍎💾 Apple is seeking US approval to purchase memory chips from China's CXMT as it looks to ease pressure from rising DRAM prices. The company is diversifying its supply chain amid an AI-driven memory shortage. If approved, the move would mark a major shift in Apple's sourcing strategy and strengthen China's position in the global memory market. #Apple | #China | #Semiconductors | #AI | #CXMT