The Week That Was: AI Provider Deals Hit $15B, EVs Get Competitive Again, and Biotech's Quiet Revolution

The pace of non-political technology news picked up meaningfully this week. AI compute deals reached stratospheric numbers, electric vehicle manufacturers started scrubbing long-standing customer complaints from their products, and biotech companies revisited foundational science in ways that could matter for decades. Here's what happened.

The AI Infrastructure Arms Race Heads to a New Level

Anthropic and the $15 Billion a Year Question

Anthropic signed a deal with SpaceX reportedly worth $15 billion per year to access AI compute capacity — and apparently, it wasn't enough. Just days later, reports confirmed that Anthropic is in early talks with Microsoft to rent servers running Microsoft's Maia 200 AI chips via Azure. It's a striking moment: Anthropic, already backed by a massive SpaceX arrangement, is still shopping for more. The inference load from Claude, particularly through its Code reasoning products and the Claude Code tooling that displaced earlier Microsoft offerings, is simply enormous.

Azure's Maia 200 chips are designed specifically to run existing large models like Claude at inference speed. Microsoft is building custom silicon as a way to reduce its own reliance on NVIDIA hardware and offer commercially differentiated AI services — and Anthropic needs this capacity for the same reason SpaceX needed what it shipped: generative AI is eating cloud infrastructure faster than anyone planned.

Anthropic's Project Glasswing Goes Public

Anthropic also formally unveiled the security tooling it developed internally for Claude Mythos Preview. Named Project Glasswing, the suite includes a skills framework, a specialized Claude harness, and a threat-model builder aimed at catching the harder-to-spot model failure modes — jailbreaks, prompt injection, and context-extraction attacks. Qualifying customers can now access these tools upon request, and Anthropic published a public dashboard of open-source vulnerability disclosures from Mythos Preview. For enterprise security teams, this closes a gap that has been a major friction point for adopting frontier models in regulated environments.

Nvidia's Numbers Are Hard to Ignore

Nvidia reported record first-quarter fiscal 2027 revenue of $81.6 billion — with data center revenue alone reaching $75.2 billion, up 92% year over year. To put that in context, this is well beyond what the entire semiconductor sector achieved even three years ago. The GPU squeeze remains severe. As rivals — including Microsoft with Maia and Google with TPUs — accelerate their own chip programs, Nvidia's throughput advantage is narrowing, but installation base, CUDA-compatible software ecosystems, and the sheer scale of global hyperscaler commitments still make the NVIDIA stack effectively irreplaceable for flagship model training today.

ChatGPT Learns PowerPoint, Gemini Learns CapCut

Two product integrations with outsized practical consequence arrived simultaneously. OpenAI added a PowerPoint plugin — a sidebar that lets users create, format, and edit entire slide decks using natural-language prompts, plus image references. Works across Word, Excel, and now PowerPoint. The toolbar integration means no more context-switching; the presentation just lives in the ChatGPT working memory alongside the documents, research notes, and images the human brought to the conversation. beta is open now for all plan tiers including free tier.

Google, meanwhile, announced that CapCut's image and video editing capabilities are coming to the Gemini app. CapCut already dominates as the free, AI-assisted video editor of choice — and Google packaging it inside Gemini means creators stay inside one interface for both drafting and editing. The underlying rationale is sound: the creative workflow is conversational and iterative, and language models that can see and modify media files are the natural execution layer for that.

AI Economics Start Looking Like Economics

Intuit announced roughly 3,000 layoffs — about 17% of its workforce — with CEO Sasan Goodarzi citing a strategic shift to AI-first services. This is becoming a pattern. Companies not investing aggressively in AI are finding that their cost structure is incompatible with the marginal-cost-equal-to-zero game that generative AI infrastructure enables. For workers, this is genuinely disruptive. For investors, the winner-takes-all dynamic is becoming unmistakable.

Aleksander Madry, one of OpenAI's most senior AI safety executives, announced he's leaving to work on something related to AI's economic impact — a credible belated signal that even inside a frontier lab, the conversation about what AI does to incentives, wages, and market structure has become urgent. The US government, while politically uncertain, introduced the Take It Down Act's platform enforcement provisions, requiring major sites to remove AI-generated non-consensual deepfakes within 48 hours of report — with actual criminal consequences for creators. Two individuals were criminally charged the day the enforcement obligations took effect, which at minimum will serve as a deterrent signal to the deepfake marketplace.

The Electric Vehicle Pivot Accelerates

Tesla Cybercab Sets a New Benchmark for Efficiency

Tesla's Cybercab has been certified at 165 watt-hours per mile — the most efficient production electric vehicle ever measured, by a wide margin. The next closest competitor, the Lucid Air Pure, uses 28% more energy per mile. VP of Vehicle Engineering Lars Moravy confirmed the certification is official and not a target figure. The practical implication: a sub-50 kWh battery in a two-seat robotaxi with no steering wheel or pedals is enough to taxi passengers reliably throughout an entire operating day without recharging, with range to spare. No other vehicle arrangement comes close on packing that amount of utility into a sub-50 kWh pack.

The asterisk is real — it only works with a purpose-built robotaxi body — but for exactly the use case it was designed for, it's a genuinely transformative efficiency number, and any competitive Fleet OEM will need to reverse-engineer it seriously.

BYD's Great Han Debut Is Imminent

BYD's flagship Great Han electric sedan was spotted testing publicly ahead of an official launch any day now. Expected specs include a 1,000 km (621 mile) WLTP range and fast charging capable of replenishing the pack in roughly five minutes. That charging time is the kind that transforms EV ownership for people without a home charger — and it directly targets the anxiety that has kept the next 40% of car buyers on the sidelines. The Great Han, sitting alongside the Great Tang SUV, positions BYD against the premium EV tier, pricing and brand permitting.

Lucid's Cosmos Is Real and It's Testing Next to a Model Y

A camouflaged Lucid Cosmos prototype was spotted testing on public roads near the Casa Grande, Arizona plant — and parked right next to a Tesla Model Y. The side-by-side test vehicle gives the first real-world size comparison for the car Lucid expects to unveil publicly in summer 2026, with production targeted for late 2026. The Cosmos is Lucid's bid for mass-market relevance. As Lucid's stock has traded near historical lows, penetrating the segment head-to-head with Tesla's volume king is make-or-break for the company's narrative.

The $20,000 EV Is Actually Arriving

Kia confirmed it is killing the Picanto — its cheapest combustion vehicle in Europe — and replacing it with an all-electric model priced around $20,000. This is the car that matters most: it compounds the supply chain scaling, charging infrastructure maturation, and battery-cost-decline dynamics that have been building for three years. A mass-market EV in that price range will not just compete with used ICE cars — it will reset the floor for what new car buyers expect to spend.

Chevy's 2027 EV Overhaul Fixes the Actual Complaints

The 2027 Chevy Equinox EV and Blazer EV are both updating to fix the package and infotainment complaints that drove down early reviews. GM has been candid about the software-electronics stack on the Bolt and early Ultium models soaking up more user patience than the hardware warranted — and for GM, software fixes are faster and cheaper than manufacturing redesigns. Getting these core models right matters enormously for Chevrolet's volume business in North America. If 2027 corrects the infotainment lag and connectivity issues that were mechanical in resolution rather than fundamental to the vehicle architecture, these products could be competitive at mainstream price points for the first time.

Biotech's Scaffolding Is Getting Rewritten

The Novartis CAR-T Anchor Moves

Novartis revisiting its T-Charge platform this week was not dramatic headline science, but it was quietly one of the most consequential platform refreshes in CAR-T biology this year. T-Charge was the technology behind Kymriah — one of the most commercially successful CAR-T approvals in oncology history — and four years later, the field has raced forward dramatically. A scaffold that made business-sense for blood cancers six years ago now looks like a starting point for the next generation of solid-tumor platforms. The competitive dynamics of neo-antigen targeting, pan-cancer T-cell receptor engineering, and in situ reprogramming are all building toward a moment when auto-immunity and cancer share the same engineering playbook.

The Circle Is Closing Between AI and Biotech

This week's convergence between AI infrastructure spending and biotech platform evolution matters because the sequencing is flipping. AI was once considered a tool for dry genomics pipelines — now it's driving antibody design, protein folding at industrial throughput, and the same kind of synthetic biology automation that used to require fleets of bench scientists. As Anthropic's compute deals make clear, the infrastructure dollar amounts are biotech-adjacent now — any organization building gravity-level AI infrastructure is effectively building a drug discovery capability whether they labeled it that or not.

What to Watch Next

The Anthropic capacity story will resolve on whether Microsoft's Maia 200 chips actually gain meaningful share against NVIDIA in high-volume AI inference — or remain a cost-reduction measure for Microsoft's own Copilot-tier deployment. The answer matters for the entire AI infrastructure value chain, from power and cooling to data center construction in the US Midwest and Mountain West, where most of the new compute capacity is being built.

On the EV side, Kia's $20,000 model launch window will be the first real test of whether a global automaker can build EVs profitably at that price point with current battery cost curves. If it can, all ICE-only MPVs and sedans in the same segment come under immediate pricing pressure.

In biotech, T-Charge revisited means foundational CAR-T assays are being reexamined — and the biotech investor community is starting to price in that solid-tumor CAR-T success is not a if but a when problem. That re-rating is in early innings.

Taken together — and keeping politics entirely out of the picture — these three vectors — AI infrastructure, EV affordability, and precision medicine — are all converging on inflection points that would have seemed distant even a year ago. The story isn't dramatic. It's actually quietly systematic, which is why it matters more than the noisy moments that don't stick.