Tech Digest – January 29, 2026

Capital & Infrastructure at Scale

Microsoft and Meta Will Spend Over $265 Billion on AI Infrastructure This Year

Microsoft’s Q2 capital expenditure hit $37.5 billion — up 66% year-over-year — with 45% of its $625 billion cloud backlog now attributed to OpenAI. The stock dropped 10% despite beating revenue estimates, as investors questioned returns on spending at this pace. Meta matched the trajectory, projecting 2026 capex between $115 billion and $135 billion, nearly double its 2025 figure, to fund what CEO Mark Zuckerberg called the company’s “superintelligence” ambitions.

Note: Institutions running workloads on Azure or planning cloud procurement should expect pricing pressure. Microsoft has already confirmed M365 price increases effective July 2026, ranging from 5% to 33% depending on the SKU. When your cloud provider spends $37.5 billion in a single quarter, the bill eventually arrives at the customer.

Sources: Bloomberg (Microsoft), Bloomberg (Meta)

OpenAI and Anthropic Reach $830 Billion and $350 Billion Valuations

OpenAI is reportedly raising $30 billion from SoftBank at an $830 billion valuation, while Anthropic is closing a $20 billion round at $350 billion. Combined, the two leading AI labs are now valued at nearly $1.2 trillion — more than most national GDPs — without traditional profitability from either company.

Note: These are not abstract numbers. These valuations set the price of the AI tools that institutions will be licensing within 12-18 months. When your future software vendor is valued like a G20 economy, understanding the business model behind the tools becomes a procurement discipline, not a curiosity.

Sources: Wall Street Journal, Financial Times

AI Chip Makers Post Record Profits as Memory Demand Outstrips Supply

SK Hynix reported a 137% surge in operating profit driven by demand for High Bandwidth Memory (HBM) chips used in AI training. Samsung tripled its quarterly profits on the same wave. Both companies are racing to expand production, but supply remains constrained — a bottleneck that affects procurement timelines for any project that depends on new compute capacity.

Sources: CNBC (SK Hynix), CNBC (Samsung)

China Greenlights 400,000 Nvidia H200 Chips for ByteDance, Alibaba, and Tencent

China has approved the import of 400,000 Nvidia H200 AI chips for its three largest tech companies. The approval signals that geopolitical barriers remain permeable when the demand for AI compute is large enough — and that chip supply is a strategic resource being managed at the state level on both sides.

Note: For European institutions watching the AI infrastructure race, this is a reminder that compute capacity is becoming a geopolitical asset. The EU’s own chip strategy under the Chips Act looks increasingly modest against the scale of what the US and China are deploying.

Sources: Reuters

Robotics & Physical AI

Tesla Discontinues Model S and X to Redirect $20 Billion Toward Robotics

Tesla announced plans to discontinue its Model S and Model X vehicles to dedicate factory capacity and an estimated $20 billion in resources to Optimus, its humanoid robot program, and broader AI infrastructure. The decision signals a corporate-level bet that robotics and autonomy will generate more value than premium vehicle sales.

Note: When one of the world’s largest manufacturers pivots from consumer products to humanoid robots, it is no longer a research curiosity — it is an industrial strategy. Institutions responsible for workforce development, manufacturing policy, or logistics should be tracking this shift, not as a future scenario but as a current capital allocation decision.

Sources: Bloomberg

Figure’s Humanoid Robot Completes a Full Dishwasher Cycle Without Intervention

Figure AI unveiled Helix 02, a Vision-Language-Action model that enabled its humanoid robot to unload a dishwasher autonomously in a four-minute, end-to-end demonstration. The robot identified items, grasped them, and placed them in the correct locations without human guidance or pre-programmed routines.

Note: A dishwasher is not the point. The point is general-purpose manipulation in unstructured environments — the same capability needed for warehouse logistics, facilities maintenance, and service delivery. The gap between “research demo” and “deployable system” is closing faster than most workforce planning models assume.

Sources: Figure AI

Science Under Compression

Anthropic Co-Founder Puts 50% Odds on AI Replacing Theoretical Physicists Within Three Years

Jared Kaplan, Anthropic co-founder and former theoretical physicist, stated he sees a 50% probability that AI will replace theoretical physicists within three years. The claim was noted by science journalist Natalie Wolchover, who added that Kaplan’s background as a published physicist gives the prediction more weight than typical Silicon Valley forecasting. OpenAI simultaneously released Prism, a free AI-native workspace for scientists to write and collaborate on research.

Note: This is not a generic “AI will change everything” claim. This is a domain expert who left physics because he believed AI would subsume it, putting a specific timeline and probability on that belief. For research institutions and education ministries, the planning question is not whether AI changes scientific work — it is whether curricula and hiring pipelines designed today will still make sense in 2029.

Sources: Natalie Wolchover (science journalist, via X), Kevin Weil, OpenAI CPO (via X)

DeepMind Publishes AlphaGenome — Predicting Gene Expression From Raw DNA Sequence

Google DeepMind published AlphaGenome in Nature, a foundation model that predicts gene expression directly from raw DNA sequence data. The model matched or exceeded state-of-the-art performance on 25 of 26 benchmarks. Separately, a new AI lab called Flapping Airplanes launched with $180 million in funding from GV, Sequoia, and Index Ventures to focus on increasing AI sample efficiency by a factor of 100,000 to 1,000,000.

Note: AlphaGenome is another data point in a pattern: AI is not just assisting scientific research — it is performing it. When a single model can match specialist tools across 25 benchmarks, the economics of scientific infrastructure change. Public research institutions funding narrow, tool-specific projects should consider whether foundation models will make those investments redundant before they deliver results.

Sources: Nature, Flapping Airplanes (via X)

AI Agents Enter the Browser

Google Puts Gemini 3 Agents Inside Chrome — Autonomous Browsing Is Now a Default Feature

Google integrated Gemini 3 into Chrome with “auto-browse” capability for shopping tasks and made it the default model for AI Overviews in Search. The company also introduced Agentic Vision in Gemini 3 Flash, turning static image processing into active visual investigation. Separately, Cloudflare stock surged 9% as enterprises adopted its tunnels to secure AI agent instances.

Note: When AI agents are embedded in the default browser of 3+ billion users, “agentic AI” stops being a conference talking point and becomes an interface pattern. Institutions managing public-facing websites should test how AI agents interact with their content — because citizens will increasingly arrive through an AI intermediary rather than a search results page.

Sources: Google Blog, The Verge, Barron’s

Workforce & Market Signals

Citigroup Mandates AI Prompt Training for All 175,000 Employees

Citigroup CEO Jane Fraser confirmed at Davos that the bank has mandated AI prompt-engineering training for 175,000 employees across 80 locations, calling it essential for employees to “reinvent themselves” before their roles change. The program, called “Asking Smart Questions,” uses adaptive learning that adjusts to each employee’s skill level. Citi reported its staff had entered over 6.5 million prompts into internal AI tools since the program launched.

Note: When a 175,000-person bank treats AI prompting as a mandatory skill — on par with compliance training — the signal for public institutions is clear. The question is not whether your workforce needs AI literacy, but when you mandate it and what “good enough” looks like at institutional scale.

Sources: Fortune

Pinterest Cuts 15% of Staff to Redirect Resources Toward AI

Pinterest is laying off nearly 15% of its workforce to redirect resources toward AI capabilities. The restructuring follows a broader pattern: companies are not just adding AI to existing teams — they are reorganizing entire headcounts around it.

Sources: Reuters

Investors Dump Software Company Bonds as AI Disruption Fears Spread to Fixed Income

Bond investors are reportedly selling debt issued by traditional software companies threatened by AI disruption. The sell-off marks a new phase: AI anxiety has moved beyond equity valuations into credit markets, where institutional money managers are repricing the long-term viability of established software business models.

Note: This matters for procurement. If the market is pricing in a future where legacy software vendors lose relevance, institutions locking into multi-year contracts with those vendors are taking a risk that the market itself is moving away from. Procurement teams should be asking vendors about their AI roadmap — not as a nice-to-have, but as a solvency indicator.

Sources: Bloomberg