The part­ner­ship between Intel and Microsoft spans more than three decades. It pre­dates the cloud era entire­ly, and it has only inten­si­fied as enter­prise work­loads have grown more demand­ing. Intel served as the first ref­er­ence archi­tec­ture for SAP HANA, and that deep tech­ni­cal col­lab­o­ra­tion across Intel, Microsoft, and SAP con­tin­ues to dri­ve the capa­bil­i­ties cus­tomers rely on today.

As an in-mem­o­ry data­base, SAP HANA oper­ates extreme­ly close to the hard­ware, con­sum­ing tremen­dous amounts of data in ways that demand tight opti­miza­tion at the sil­i­con lev­el. Such prox­im­i­ty makes the col­lab­o­ra­tive rela­tion­ship essen­tial rather than optional.

At hyper­scale, every effi­cien­cy gain com­pounds: every watt saved in a proces­sor mul­ti­plies across mil­lions of process­es, and every improve­ment in how sys­tems han­dle large data sets trans­lates direct­ly to cost and per­for­mance out­comes. Intel’s posi­tion in mul­ti-sock­et scal­ing, the IP required to move from two sock­ets to four, eight, or 16, remains dif­fi­cult to repli­cate, and it under­pins the large-scale SAP deploy­ments that Azure sup­ports today.

At Microsoft Ignite 2025, Azure announced vir­tu­al machines (VMs) pow­ered by Intel’s newest Xeon 6 proces­sors, includ­ing sup­port for CXL-based mem­o­ry expansion.

Three of these devel­op­ments land at a moment when SAP cus­tomers face a famil­iar set of pres­sures: in-mem­o­ry data­bas­es con­sum­ing ever-larg­er foot­prints, mis­sion-crit­i­cal sys­tems that can­not tol­er­ate down­time, and finance teams scru­ti­niz­ing every dol­lar spent on infrastructure.

1. CXL Tech­nol­o­gy Comes to the Cloud

A decade ago, run­ning a VM with 32 ter­abytes of capac­i­ty and more than 1,700 vir­tu­al CPUs in a hyper­scale cloud would have seemed far-fetched. Today, Azure sup­ports VMs at that scale.

Get­ting there required sus­tained invest­ment across all three part­ners, opti­miz­ing vir­tu­al­iza­tion lay­ers and ensur­ing reli­a­bil­i­ty across an enor­mous num­ber of com­po­nents. The new Xeon 6 proces­sors push the ceil­ing high­er still. Intel’s ⁶⁵⁰⁰⁄₆₇₀₀ series, announced this year, sup­ports scale-up con­fig­u­ra­tions reach­ing up to 64 terabytes.

Equal­ly sig­nif­i­cant is the intro­duc­tion of CXL Flat Mem­o­ry Mode, the first hard­ware-man­aged mem­o­ry tier­ing sys­tem for CXL mem­o­ry, now avail­able in pri­vate pre­view on Azure. CXL tech­nol­o­gy decou­ples capac­i­ty scal­ing from com­pute scal­ing, allow­ing orga­ni­za­tions to tai­lor con­fig­u­ra­tions to actu­al work­load require­ments rather than over­spend­ing on com­pute sim­ply to access the resources they need. For large S/4HANA deploy­ments, this adds new flex­i­bil­i­ty in how mem­o­ry resources can be con­fig­ured. All this with­out com­pro­mis­ing per­for­mance, thanks to the unique capa­bil­i­ties of Intel Xeon 6 and Intel Flat Mem­o­ry Mode. 

As the first hyper­scaler to have announced such capa­bil­i­ty, Microsoft Azure con­tin­ues its strong his­to­ry of tech­nol­o­gy lead­er­ship with SAP infrastructure. 

2. Broad­er VM Fam­i­lies Join the Lineup

SAP cus­tomers come in all sizes, and not every work­load demands 32 ter­abytes of capac­i­ty or 1,700 vCPUs in a sin­gle VM. The Ignite announce­ments includ­ed new gen­er­al-pur­pose and mem­o­ry-opti­mized VM options based on Intel Xeon 6 proces­sors — the Dlsv7, Dsv7, and Esv7 series — now in preview.

These deliv­er com­pute per­for­mance improve­ments of up to 15% com­pared to pre­vi­ous-gen­er­a­tion Intel-based VMs, with the largest con­fig­u­ra­tions scal­ing to 372 vCPUs and 2.8 ter­abytes of capac­i­ty. Net­work­ing band­width reach­es 400 giga­bits per sec­ond; remote stor­age through­put hits 800,000 IOPS.

The new VMs sup­port three capac­i­ty-to-vCPU ratios, with options includ­ing and exclud­ing local NVMe temp disks. Gen­er­al-pur­pose Dlsv7 and Dsv7 con­fig­u­ra­tions suit a range of com­mon work­loads, from web-fac­ing appli­ca­tions to back­end appli­ca­tion servers. Mean­while, the capac­i­ty-opti­mized Esv7 fam­i­ly tar­gets work­loads with high­er demands: large-scale data­bas­es (both SQL and NoSQL), ana­lyt­ics and busi­ness intel­li­gence plat­forms, and in-mem­o­ry sys­tems like SAP S/4HANA.

3. Con­fi­den­tial Com­put­ing for Reg­u­lat­ed Workloads

The col­lab­o­ra­tion between Intel and Microsoft extends to the secu­ri­ty fea­tures built into the sil­i­con itself, includ­ing trans­par­ent encryp­tion and hard­ware-lev­el iso­la­tion. Azure’s next-gen­er­a­tion con­fi­den­tial VMs rep­re­sent the lat­est devel­op­ment in that realm.

Now in pre­view with gen­er­al avail­abil­i­ty expect­ed in Q1 2026, these VMs are pow­ered by fifth-gen­er­a­tion Intel Xeon proces­sors with Intel Trust Domain Exten­sions (TDX), enabling orga­ni­za­tions to run con­fi­den­tial work­loads with­out mod­i­fy­ing appli­ca­tion code. Data stays encrypt­ed even dur­ing pro­cess­ing, address­ing what had been a gap in end-to-end protection.

For cus­tomers in health­care, defense, finan­cial ser­vices, and oth­er reg­u­lat­ed indus­tries, con­fi­den­tial VMs pro­vide iso­la­tion enforced at the sil­i­con lev­el, keep­ing sen­si­tive code and data encrypt­ed while in use. Ear­ly adopters have deployed TDX-enabled VMs for secure col­lab­o­ra­tion plat­forms, end-to-end encryp­tion archi­tec­tures, and con­fi­den­tial AI envi­ron­ments that meet strict data sov­er­eign­ty requirements.

The con­fi­den­tial VM offer­ings also include Intel Advanced Matrix Exten­sions (AMX), an accel­er­a­tor built into the proces­sor. AMX sets the foun­da­tion for emerg­ing AI work­loads that require hard­ware accel­er­a­tion while main­tain­ing con­fi­den­tial­i­ty guar­an­tees. Upcom­ing releas­es will add local NVMe SSD sup­port, which allows for approx­i­mate­ly five times the stor­age through­put and low­er laten­cy com­pared to the pre­vi­ous SCSI generation.

The Part­ner­ship Behind the Products

The announce­ments at Microsoft Ignite reflect years of joint devel­op­ment, but the partnership’s val­ue to cus­tomers extends beyond any sin­gle prod­uct release.

Reli­a­bil­i­ty is a case in point. The dif­fer­ence between man­ag­ing a serv­er with half a ter­abyte of capac­i­ty and one with more than 40 ter­abytes is more than only quan­ti­ta­tive. At that scale, the sheer num­ber of DIMM mod­ules in play means error cor­rec­tion and fault com­pen­sa­tion move from edge cas­es to cen­tral design concerns. 

Intel has invest­ed heav­i­ly in reli­a­bil­i­ty, avail­abil­i­ty, and ser­vice­abil­i­ty capa­bil­i­ties across Xeon proces­sor gen­er­a­tions, and the col­lab­o­ra­tion on test­ing and val­i­da­tion, tun­ing per­for­mance, stress-test­ing fail­ure sce­nar­ios, and cer­ti­fy­ing con­fig­u­ra­tions starts years before any proces­sor reach­es gen­er­al availability. 

Those fun­da­men­tals show up in cus­tomer out­comes. Migra­tions to Azure have pro­duced per­for­mance increas­es of 30% to 40% across many sys­tems, with ana­lyt­ics and busi­ness ware­house work­loads improv­ing by as much as 120% com­pared to old­er on-prem infrastructure.

Orga­ni­za­tions that once main­tained sep­a­rate SAP sys­tems by geog­ra­phy or busi­ness unit — often because infra­struc­ture con­straints made con­sol­i­da­tion imprac­ti­cal — are now mov­ing toward glob­al S/4HANA instances. Reduc­ing the num­ber of sys­tems under man­age­ment gen­er­ates dra­mat­ic cost savings.

Equal­ly impor­tant is the assur­ance that growth paths exist. Com­mit­ting to an archi­tec­ture is eas­i­er when dou­bling busi­ness vol­ume that does not require rad­i­cal sys­tem changes. The sus­tained invest­ment across Intel, Microsoft, and SAP means cus­tomers can plan for expan­sion know­ing the infra­struc­ture will keep pace with the needs of the business.