AI Memory Investment, Apple Price Increases, IBM 0.7nm and Physical AI
July 21, 2026
Every semiconductor announcement has two stories—the one on the surface, and the real story grounded in physical evidence, market data, and 35 years of reverse engineering. This is where you find the second one.
Samsung and SK hynix Announce Memory Fab Investments
Samsung and SK hynix are making massive new investments in memory manufacturing, but the real story is what those decisions signal. Their expansion plans reflect growing confidence that AI-driven HBM demand is fundamentally changing the economics of the memory industry.
On June 24, 2026, SK hynix filed to list American depositary receipts (ADRs) on the Nasdaq, targeting approximately $29 billion (45.45 trillion won) through the issuance of 17.79 million new shares. Five days later SK hynix published its mid-to-long-term investment strategy: a KRW 1,100 trillion (~$714 billion) commitment across three South Korean clusters.
Separately, Samsung announced plans for a new chip complex in Gwangju as part of its own parallel expansion, alongside investments in HBM fabs in Cheonan and Onyang, alongside existing semiconductor back-end processing fabs, physically locating HBM memory production with its existing packaging and stacking technology investments.
The Nasdaq ADR listing is a market referendum on whether investors accept that AI-driven memory demand is structurally different from the commodity cycles that defined the industry for decades. TechInsights data supports the structural case. HBM bit shipments are forecast to grow at a 94% CAGR, a sustained appreciation that has no precedent in the commodity era. SK hynix's HBM production is sold out through 2026, and every bit of HBM produced consumes the equivalent of more than three bits of conventional DRAM wafer capacity, meaning supply cannot quickly catch demand even with aggressive capex.
For the broader South Korean semiconductor ecosystem, Samsung's parallel announcement means two of the world's three largest DRAM producers are simultaneously competing for construction labor, equipment, electricity, and water infrastructure in South Korea.
Apple Increases Consumer Prices
Apple confirmed device price increases are coming, but the real story is what's forcing them. Memory makers have redirected DRAM and NAND capacity toward AI infrastructure, leaving consumer electronics starved for supply. Apple can absorb the hit. Other device makers may not be so lucky.
Apple CEO Tim Cook confirmed that price increases on Apple devices are unavoidable, citing unprecedented DRAM and NAND memory cost inflation driven by AI infrastructure demand. Cook described the situation as unlike anything he had experienced in over four decades in the electronics supply chain. The confirmation ends any ambiguity: higher MSRPs are coming across almost all Apple devices.
The root cause is structural. The three major memory manufacturers, Samsung, SK hynix and Micron, have shifted production capacity toward high-bandwidth memory for AI data centres, leaving consumer-grade DRAM and NAND in persistent short supply. Industry contract pricing data shows costs rising sharply through H1 2026, with no meaningful new supply expected before late 2027 at the earliest.
Component and manufacturing costs for a base Pro model are tracking roughly 25% higher than last generation, according to TechInsights, and are driven almost entirely by increased memory pricing, The numbers leave Apple no option. Apple is the highest profile announcement so far, but other manufacturers will also have to follow suit. Apple’s brand and software ecosystem mean that it is likely to be able to continue to attract consumers even at a higher price point. Other vendors may face more significant consumer resistance to price increases.
IBM Debuts Sub-1nm Nanostack Chip Technology at 0.7nm
IBM unveiled a 0.7nm chip built on a new nanostack transistor architecture, but the real story is what it does to the roadmap. Vertical integration just proved it can extend scaling well past the limits everyone assumed for planar designs. The open question is which foundry picks it up, and when.
IBM unveiled the world's first sub-1 nanometer chip technology, built on a new transistor architecture called "nanostack" at the 0.7nm (7 angstrom) node. The chip packs nearly 100 billion transistors onto a die the size of a fingernail, roughly twice the density of IBM's 2nm nanosheet chip first demonstrated in 2021. IBM projects the technology will deliver up to 50% more performance or 70% greater energy efficiency versus its 2nm node, based on results presented at VLSI 2026.
IBM's new technology confirms that vertical 3D integration of nanosheet stacks can sustain density scaling past the physical limits of planar gate-all-around (GAA) designs. The nanostack work extends the roadmap for continued transistor scaling by at least a decade, a runway that came to a worrying and abrupt halt in 2030. This extends semiconductor roadmap visibility out to somewhere between 2040 and 2050.
Rapidus, which is commercializing IBM's 2nm GAA technology with volume production targeted for 2027 (6 years after its introduction), provides the nearest production reference point for IBM's research lineage. Whether a foundry partner picks up 0.7nm for production will depend on High NA EUV tool availability and process maturity, both of which remain early-stage.
onsemi acquires Synaptics for $7 billion
onsemi is acquiring Synaptics for $7 billion, and Groq just raised $650 million to rebuild, but the real story is how differently that AI money is moving. One deal buys into Physical AI before NVIDIA gets there first. The other is what's left after NVIDIA already took the best parts.
onsemi and Synaptics Incorporated announced that they have entered into a definitive agreement under which onsemi will acquire Synaptics in an all-stock transaction, representing a total enterprise value of approximately $7 billion. The combination would accelerate onsemi’s evolution toward global leadership in intelligent systems. By adding Synaptics’ differentiated Edge AI compute franchise and a strong portfolio of human-machine interface and wireless connectivity solutions, onsemi is expected to extend its capabilities beyond power and sensing into intelligent systems.
onsemi’s planned acquisition of Synaptics is a strategic move to expand beyond power and sensing into broader intelligent edge systems. By adding Synaptics’ Edge AI compute, connectivity, and human-machine interface portfolio, onsemi aims to strengthen its position in Physical AI: systems that can sense, decide, act, and adapt in real time. NVIDIA has already begun its push into this new era, which shifts AI from the datacenter to real-world environments.
Groq raises $650 million
Groq is repositioning itself as an AI inference cloud services provider after NVIDIA acquired much of its IP and many key employees. The new capital supports the expansion of Groq's global data center footprint and its goal of scaling toward 200 MW of compute by the end of 2027.
Groq today announced $650 million in new growth capital to accelerate the expansion of its AI inference cloud. The round was led by Disruptive and Infinitum, with participation from investors who elected to reinvest in the company. Coincident with the funding, Groq also announced multiple new additions to its leadership team. Alan Rice (previously of xAI and Meta) joins as COO. Sinclair Schuller and Rakesh Malhotra, both previously of Apprenda and then Nuvalence, join as CTO and CPO (Chief Product Officer) respectively.
The overwhelming majority of Groq, including much of its IP along with many of its key staff, were “aqui-hired” by NVIDIA for $20 billion at the tail end of last year. Groq is now repositioning itself as a AI cloud services provider focused on inference, with 13 datacenters globally. The new funding demonstrates that investors back the new direction for what remains of Groq after NVIDIA. The new leadership team is a much-needed addition, after the majority of the old leadership team left to join NVIDIA. The new capital will accelerate the fit-out of its existing footprint with the new LPX system from NVIDIA, the IP for which is a significant part of what NVIDIA acquired. Groq expects to scale toward 200 MW of compute by the end of 2027. This new funding will, in effect, mostly benefit NVIDIA once again.
Get the evidence behind the semiconductor story.
Semiconductor news moves fast. Understanding what it means requires more than following the headlines. Get Behind the Headlines delivered right to your inbox and discover the semiconductor stories worth understanding.
Just register with your corporate email. No purchase required to access the free insights highlighted in this edition.
