Meta is showing off the custom hardware it's building at its Menlo Park lab to power next-generation AI systems. The piece features developer Tom Shaw walking through what the company is actually constructing behind the scenes.

• Meta is developing proprietary hardware specifically designed for AI workloads rather than relying solely on off-the-shelf chips
• The Infrastructure Lab is located in Menlo Park and serves as Meta's central hub for hardware R&D
• Custom infrastructure is key to Meta's strategy for personalization, ad targeting, and content moderation at scale

Etched announced a working inference chip and over $1 billion in signed contracts after emerging from stealth with $800 million raised and a $5 billion valuation. Their rack-scale system, built on TSMC’s N4P process, is already running models like DeepSeek, Qwen and Llama, and production is ramping at new Taiwan and San Jose facilities.

• Etched has working A0 silicon on TSMC's N4P process and a rack-scale system already running DeepSeek, Qwen, Mamba and Llama models
• The company raised $800M total, including a $500M round in December at a $5B post-money valuation, backed by Peter Thiel, Karpathy, Hinton, Fei-Fei Li and major trading firms
• It has over $1 billion in signed customer contracts despite being just under three years old
• Etched is building out Taiwan production and a San Jose HQ (data center, test house, NPI lab) with a goal of gigawatt-scale manufacturing by 2027

Sunrun, Tesla and Renew Home plan to pool home batteries and smart thermostats into a 16 GW virtual power plant aimed at supplying data centers facing fast-growing electricity demand. They say this “capacity-as-a-solution” can be deployed in months with no new infrastructure, and have already lined up 300 MW in Virginia’s Data Center Alley. The move targets hyperscalers racing to power AI loads and could cut grid costs by shaving peak demand rather than building new plants.

• Sunrun, Tesla, and Renew Home are pooling 16.8 GW of home batteries and smart thermostats into a VPP to power data centers, but only 300 MW in Virginia is actually live/dispatchable today—the rest is aspirational or based on mixed metrics (rated capacity vs. one-hour load-shift potential).
• The pitch is speed: this can deploy in months via software, versus years for new transmission lines or power plants, targeting hyperscalers who can't wait for traditional grid buildout.
• A Brattle Group study backing the plan claims better use of existing grid assets (shaving peak demand instead of building new infrastructure) could save $110–170 billion in U.S. power costs over a decade and speed up data center grid hookups.
• Success depends entirely on unproven variables—customer enrollment, utility approvals, grid-operator acceptance, and locking in actual purchase contracts with data center operators.

This article breaks down the massive debt and revenue milestones that AI leaders (NVIDIA, OpenAI, Anthropic) must hit to justify the $9–15 trillion in planned data-center build-out. It shows how banks, hyperscalers, and chipmakers need AI services to generate over $2 trillion annually by 2030 or risk a market collapse.

• Building the planned 190 GW of AI data-center capacity could cost $9.5–15 trillion (far above Bloomberg's $3 trillion estimate), requiring banks to roughly double annual debt issuance to $500B–$1T just to sustain it
• NVIDIA's projected $1 trillion 2027 revenue depends heavily on three clients (likely ODMs for Microsoft, Google, Meta), tying its fate to those firms' ability to keep raising debt
• OpenAI and Anthropic will drive 70–90% of AI compute demand but are on track for under $360 billion combined revenue by 2029—less than half the ~$875 billion needed even under a scenario where only half the planned capacity gets built
• Outside the major AI labs there are essentially no other large-scale compute buyers, meaning enterprise IT spending on AI would need to grow by orders of magnitude to justify current valuations and debt levels

This article explores the massive energy consumption of AI data centers, particularly focusing on Elon Musk's Colossus facility in Memphis. It highlights the reliance on fossil fuels, the environmental impact, and the rapid expansion of data centers across the U.S. as companies race to develop advanced AI models.

• Musk's Colossus data center in Memphis will use as much electricity annually as the entire city of Seattle, powered by its own natural-gas plant with up to 35 turbines.
• Data centers' energy use could surpass that of all U.S. heavy industries combined by 2030, with the IEA projecting emissions from them could more than double.
• Colossus alone consumed over 11 million gallons of water in a single month for cooling.
• Loudoun County, VA already hosts 199 data centers, illustrating the rapid nationwide buildout also underway in Phoenix, Atlanta, and Dallas.

Microsoft CFO Amy Hood halted some data center projects after realizing the company was overspending on infrastructure for AI and cloud services. This move comes amid concerns about a potential tech bubble as the company navigates rising costs and demand.

• Amy Hood paused some data center projects in late 2024 after spotting Microsoft was set to spend more in a single quarter than in all of the prior year
• Her intervention highlights growing internal concern at Microsoft about overspending amid fears of an AI/tech bubble
• Microsoft still faces pressure to keep building AI/cloud infrastructure fast enough to stay competitive despite this financial caution
• Hood's decisions are positioned as pivotal to whether Microsoft's AI bet pays off without overextending the company financially