AI Infrastructure and Compute
The hardware and cloud stack behind AI, including GPUs, accelerators, data centers, networking, inference chips, and compute supply constraints.
Modal Has Run More Than 3 Billion Sandboxes for AI Agents
Modal CEO Erik Bernhardsson argues that the new capabilities of AI models require new infrastructure, including inference, training and sandboxes that let agents execute code in isolation. He describes those tools as building blocks for AI applications; Modal says it has run more than 3 billion customer sandboxes. The company has raised $750 million at a $15.75 billion post-money valuation to grow faster, launch products and add capacity.
Meta’s Frontier AI Claim Raises Questions Beyond Benchmark Scores
John Coogan said *The Social Reckoning* feels dated because it revisits controversies that no longer capture the questions technology professionals are asking about Meta. He argued that a current account would scrutinize the company’s AI business, including whether Muse Wonder’s benchmark results establish the frontier capabilities Meta claims. Jordi Hays was impressed by Muse as a product but said the benchmarks alone did not convince him Meta had reached the frontier.
Huang Says Export Controls Must Protect Technologies Without Shrinking U.S. Platforms
In his Taiwan travel video with Lex Fridman, NVIDIA chief executive Jensen Huang argues that technological leadership depends on more than controlling scarce capabilities: it also requires the customers, partners and industrial networks that help a technology platform grow. He applies that logic to export controls, making the case for selectively restricting technologies while preserving the reach that sustains innovation and prosperity. His account of Taiwan’s semiconductor ecosystem shows why, in his view, that reach depends on international relationships as well as domestic investment.
Taiwan’s Alternatives to Chinese Suppliers Still Depend on Chinese Inputs
Taiwan’s supply-chain exposure extends beyond advanced chips: moving production away from China does not necessarily remove dependence on Chinese-controlled suppliers operating abroad or on Chinese feedstock and processing. In a Hoover Institution roundtable, researchers from Taiwan’s Democracy, Society and Emerging Technology institute argue that resilience depends on building and qualifying alternatives without disrupting production before those substitutes are ready.
Microsoft Plans Multi-Gigawatt Maia Deployment for Internal AI Workloads
Microsoft is moving its Maia chips into data centers as it seeks to run more of its AI workloads on homegrown hardware. Bloomberg reporter Matt Day said the company plans a multi-gigawatt deployment, with its third-generation Maia 300 beginning rollout in 2027, and expects internal services such as Copilot and GitHub to use the capacity. Microsoft’s case for the chips is that they can run some workloads more efficiently and reduce the cost of its data-center operations.
Oracle Trucks Natural Gas to Keep Data Center Builds on Schedule
Oracle is trucking natural gas to data centers near Salt Lake City and is considering the same approach for a New Mexico site where a needed pipeline is delayed. Bloomberg’s Brody Ford described trucked gas as a stopgap, not a preferred long-term power source: the workaround reflects pressure to bring data centers online quickly after Oracle has borrowed heavily to build them.
Hyperscaler Spending Can Slow While Still Outpacing Economic Growth
Goldman Sachs Wealth Management’s Matt Weir argues that the market’s reliance on a handful of large technology stocks does not, on its own, signal that the bull market is ending. He expects hyperscalers’ capital-spending growth to slow sharply from its extraordinary pace, but remain far faster than economic growth. For Weir, earnings—not short-term market concentration or macroeconomic distractions—will ultimately determine where stocks go next.
Samsung’s Record Profit Trails Estimates as Memory Demand Outpaces Supply
Samsung reported record operating profit of 107.4 trillion won for the three months through September, as rising memory-chip prices and AI infrastructure spending lifted earnings. CLSA Korea Research Head Sanjeev Rana said the result was broadly in line with his expectations, despite falling short of the analyst estimate shown with the figures. He remains bullish on the memory cycle, citing demand that exceeds supply and expecting a sharp rise in high-bandwidth memory pricing in 2025.
SpaceX’s Chip Financing Talks Put AI Debt Appetite to the Test
SpaceX is in early-stage talks to raise $40 billion to buy Nvidia chips, Bloomberg Tech reporter Ed Ludlow said, in a financing that could include $10 billion in loans and $30 billion in investment-grade credit. Ludlow said the proposal highlights questions about how investors will finance the AI buildout, including how debt fits the useful economic life of GPUs and whether companies may borrow heavily after going public.
AI Excels at Verifiable Work, While Building Still Depends on Human Judgment
OpenAI’s mathematical results illustrate the rapid progress John Coogan sees in tasks that can be carried out and checked in digital environments, but he argues that technical capability does not settle the broader work of building a company. That work still involves judgment, hiring, dealmaking and execution. In the discussion of SpaceX’s proposed $40 billion chip financing, Jordi Hays argued that investors must assess the risks and valuation themselves, even if Elon Musk’s track record earns him some latitude.
Goodput, Not FLOPS, Measures What Frontier AI Systems Actually Deliver
Amin Vahdat, Google’s chief technologist for AI infrastructure, argues that theoretical chip speed is a poor measure of AI capacity: at large scale, failures and recovery determine how much useful work a system delivers. He calls that measure goodput, and says it should guide decisions across the infrastructure stack, from chip design and networking to power and storage. As AI workloads shift toward long-running agents, Vahdat argues, the systems supporting them must evolve beyond accelerators alone.
Anthropic’s $518 Billion Compute Commitments Face Uncertain Customer Demand
Anthropic’s leaked financial figures show revenue rising from $400 million in 2024 to $4.6 billion in 2025, alongside a decade-long commitment to $518 billion in computing and infrastructure spending. On Big Technology Podcast, Alex Kantrowitz argues that those obligations could amplify the effects of slower growth, both for Anthropic’s suppliers and for a stock market increasingly reliant on AI-related companies. Ranjan Roy sees customer concentration and losses as more immediate risks, and expects Anthropic’s counterparties to have reasons to negotiate if the company falters.
Nvidia’s Cloud Strategy Shifted From GPU Marketplace to Infrastructure Software
Nvidia’s DGX Cloud Lepton began as a bid to give developers a single front door to GPU capacity, potentially putting the company in competition with major cloud providers. John Coogan recalled that Jensen Huang framed the opportunity as a way for Nvidia to compete with those providers; Coogan argues that Lepton instead evolved into software for managing AI infrastructure across them, extending Nvidia’s reach without requiring it to become the cloud. The article also examines separate debates over executive conduct in the Factory–Cognition dispute and the role of shopping recommendations in AI assistants.
Saronic Bets $3 Billion on a New Shipyard to Scale Autonomous Shipbuilding
Saronic CEO and co-founder Dino Mavrookas argues that the United States needs to expand shipbuilding capacity quickly as China invests in the industry. The company is committing more than $3 billion to Port Alpha, a new Brownsville, Texas, shipyard it expects to open by the end of 2027, with its first ships due a year later. Mavrookas says building capacity ahead of demand is essential to scaling production of autonomous vessels.
Cambridge Proposal Calls for Comparable AI R&D Metrics and Crisis Plans
Government contact with AI lab leaders is increasing as Cambridge researchers call for comparable measures of AI’s role in research and development and plans for severe failures. TBPN hosts John Coogan and Jordi Hays also raise a specific market risk: if widely used agents make similar financial decisions at the same time, they could contribute to a bank run or flash crash. Coogan cautions that adoption and trust will take time.
AI’s Rapid Progress Is Making Corporate Dominance Harder to Predict
In a conversation with Shane Parrish, Pershing Square founder Bill Ackman argues that rapid advances in AI make it harder to know whether today’s dominant companies will remain durable: some, he says, may disappear. The investment case he lays out is to scrutinize a business’s long-term resilience, management and capacity to generate cash, while waiting for a price that offers an attractive return. He also describes how a change in Netflix’s plans broke Pershing Square’s investment thesis—and why evidence of stronger execution later led the firm to buy the stock again.
Microsoft’s Autopilot Bet Puts Consumer Distribution at the Center of AI Agents
Microsoft is recasting Copilot as Autopilot, a consumer agent designed to work across a user’s services and personal life. On TBPN, John Coogan and Jordi Hays argue that Microsoft’s reach among ordinary consumers could matter as much as the agent’s capabilities: people may encounter it through tools they already use, even if AI enthusiasts are not drawn to a Microsoft-branded product. But acting across services asks users to trust an agent in a more active role.
Trump-Xi Summit Produced Favorable Optics but Little Visible Progress
Julian Gewirtz, a former senior director for China and Taiwan Affairs at the White House, says Donald Trump’s meeting with Xi Jinping achieved its apparent aim of projecting a positive relationship but produced little visible progress on the disputes dividing the United States and China. He argues that trade, Iran, Russia and Taiwan remain unresolved, while competition over AI and access to advanced chips adds to the stakes.
Meta’s Consumer AI Strategy Puts Integration Ahead of Benchmark Leadership
John Coogan argues that Meta’s consumer-AI strategy is becoming clearer: rather than win every model benchmark, the company can pair capable agents with its infrastructure, hardware and reach. He sees Muse as a practical assistant for ordinary tasks, while Jordi Hays questions whether Meta’s devices will keep users beyond their initial appeal.
Physical AI Could Multiply Global GDP, but Depends on Rebuilding Manufacturing
Morgan Stanley analyst Adam Jonas argues that combining AI with machines operating in the physical world could multiply global GDP by eight to ten times, but says that prospect depends on rebuilding manufacturing capacity. He sees China’s strength across key supply chains as a constraint on US efforts to onshore production, and expects physical AI to develop through networks of connected machines rather than humanoid robots alone.
Boom Plans to Finance Supersonic Flight With Data Center Power
Boom Supersonic founder Blake Scholl argues that the company’s engine program can finance and de-risk its planned Mach 1.7 airliner by first supplying trailerized, behind-the-meter power to AI data centers. After losing its intended engine supplier, Boom chose to develop and manufacture its own turbine core; Scholl says that same core can generate 42 megawatts in a stationary configuration while accumulating operating time before it flies on Overture. His broader case is that vertically integrated manufacturing, boomless overland cruise and a roughly $3,500 transatlantic break-even fare can make commercial supersonic service viable again.
FlashNorm Folds RMSNorm Into Projection Weights to Cut Decode Time
Filip Makraduli argues that RMSNorm’s inference cost comes less from arithmetic than from repeated GPU launches, memory movement and dependencies that hold up projection matmuls. His FlashNorm approach exactly rewrites the norm–projection sequence by folding learned gains into weights, deferring the RMS scalar division until after the matmul, and in some architectures removing a redundant normalization. Weight folding can speed existing checkpoints now, he says, while the larger benefit from deferred normalization depends on fused kernels and correct CUDA-stream synchronization.
Hugging Face Bets on Cheap Robots to Build Physical AI
Hugging Face co-founder Thomas Wolf argues that Microduck, a sub-$400 open-source robot, is meant to bring more builders into physical AI—not by solving household work, but by giving users a capable machine on which to train and share behaviors. In this interview, Wolf places that bet alongside Hugging Face’s larger model: free public distribution supported by paid storage and compute infrastructure, with open models providing organizations an alternative when proprietary systems are unavailable, restricted or unsuitable. He also says that openness must be paired with stronger AI-driven cyber defense after the company’s own breach.
Crusoe Raises $3.9 Billion to Expand Its AI Infrastructure Stack
Crusoe CEO Chase Lochmiller says the company’s $3.9 billion funding round will support an AI infrastructure business that spans long-term data-center contracts, managed GPU clusters and on-demand AI services. He argues that the sector’s growing local opposition reflects a “marketing issue,” contending that purpose-built facilities can limit water use while generating substantial tax revenue for host communities.